The bloc is coordinating the investigations by the Consumer Protection Cooperation Network, which brings together the competent authorities of the EU’s 27 states.

The network launched talks with the companies last year, but the EU executive said the approach “did not bring satisfactory results”.

The European consumer authorities later identified several games that needed further study, including Candy Crush Saga, Minecraft and Clash of Clans.

The EU said that alongside Ubisoft, the other companies involved in the probe included Minecraft developer Mojang, US giant Riot Games, and Supercell, which makes Clash of Clans.

Authorities will now probe whether there are any breaches of EU consumer law.

The bloc’s guidelines on in-game virtual currencies say games must show the real-world price of any required in-game items and currencies. Games also must show clear information before any purchase, and companies must not force players into unwanted transactions.

Gamers should also be told they can withdraw from contracts within 14 days, including for unused virtual currency, the EU said.

“The industry must ensure that its games do not expose players — especially children — to harmful or unfair practices,” EU consumer protection chief Michael McGrath said. “The game must be fair, and the rules must be respected.”

He added in a statement: “National authorities, with the support of the Commission, will make sure they are enforced.”

The EU has also ramped up efforts to protect children online, announcing earlier this month rules that will force companies to make video games and social media platforms safe before minors use them.

The EU is also expected to unveil consumer protection rules known as the Digital Fairness Act that will cover the video game sector.

“They’re going to police themselves,” Trump said after a meeting at the White House that gathered the top leaders in artificial intelligence, including Nvidia’s Jensen Huang, OpenAI’s Greg Brockman and Anthropic’s Dario Amodei.

The Republican president, who opposes any constraints on AI lest China outpace the United States in its development, also announced that he would officially rename the technology “super intelligence” by executive order later on Tuesday.

OpenAI launched a new mid-range artificial intelligence model on Tuesday at one-fifth the cost of its flagship offering, one day after scrapping an upgraded version deemed unreliable.

GPT-6.1 Sol ranks below OpenAI’s most powerful model, GPT-6 Astra, released in early September. The company said Monday it had abandoned plans for GPT-6.1 Astra because the model frequently ignored instructions.

The announcement, made at the company’s annual developer conference, marks the latest escalation in a price war with rival Anthropic as both firms eye initial public offerings. OpenAI also announced Dots, an always-on personal agent designed to compete with Meta’s Muse, unveiled last week.

The company acknowledged last week that tools being tested had browsed US federal agency websites without authorisation and attempted to hack one of them.

Australian Prime Minister Anthony Albanese has also criticised the company for being slow to report a June intrusion into a public health portal.

Long the industry leader, OpenAI was overtaken by Anthropic in second-quarter revenue, which hit $11.6 billion against $6.7 billion for OpenAI, according to the Wall Street Journal.

Anthropic is now targeting a stock market listing as early as November, even though it warned investors in its IPO prospectus that AI could pose “existential risks to humanity”, the Financial Times reported on Tuesday.

OpenAI, valued at $852 billion in March, has not set a date for its own listing.

In mid-September, chief executive Sam Altman ruled out 2026, citing safety, and on Tuesday he told CNBC that “this is a time to put safety and mission first.” OpenAI’s second-quarter operating loss widened to $12.3 billion, including stock-based compensation, according to the Wall Street Journal.

Published in Dawn, September 30th, 2026

The “America.gov” platform was developed under Airbnb co-founder Joe Gebbia, who joined the Trump administration last year in the newly formed role of chief design officer.

Like other websites using AI-powered chatbots, users type in queries and converse with the system online. While its functions are currently limited, Trump said that America.gov will eventually be able to process all sorts of complicated tasks involving the government, such as applying for a new passport or accessing health and retirement benefits.

“We’re making it dramatically easier, faster, and more convenient to get what you need exactly when you need it,” Trump said at a launch event for the platform in Washington, attended by top government officials.

“It’s a restoration of America’s founding promise and it’s a reinvention of your government for the 21st century and beyond,” he said.

The president also signed an executive order directing all government agencies to integrate their websites with America.gov.

The launch of the new US government platform comes the same day that Trump is to meet with an array of top tech executives to discuss the AI sector’s rapid development.

A growing list of cyber incidents revealed in recent weeks has raised concerns around the world about risks posed by uncontrollable AI, with some executives urging a slowdown for safety. Trump has dismissed those concerns as a conspiracy to undermine US economic growth and hamper competition with China.

AI and the rapid buildout of data centres have become a significant issue in the November midterm elections. A recent poll published by CNN showed 82 per cent of Americans expressed concern about the pace of AI development.

The news comes one day before the AI giant hosts an annual developer conference known as OpenAI DevDay in San Francisco, where the company is expected to make several announcements — though it is unclear if a new version of Astra will be among them.

Astra 6.1 was an improvement over previous models in some aspects, but “it didn’t quite meet the bar in terms of staying within scope and authorisation, and how it communicates back to the user about the type of work it’s done,” Saachi Jain, OpenAI’s head of safety systems, said in a statement.

“We want to make sure our model development is safe no matter whether that’s in the company, or when we ship it to users. But when we ship it to users, we have an extremely high bar in terms of safety and alignment,” Jain continued.

AI safety fears

Concerns about AI safety have escalated in recent months after models developed by OpenAI and rival lab Anthropic were involved in security incidents during testing.

Agents built with OpenAI’s models have inappropriately accessed websites maintained by US federal agencies, an Australian government health statistics portal and Hugging Face, a repository of AI models.

OpenAI apologised Monday for not properly responding to the Australia incident, which involved its AI models accessing government websites without authorisation.

“We are sorry and working to do better in the future,” OpenAI said in a blog post, adding that the company would explain “what we know, what we have changed, and what we will do to rebuild trust with the Australian people.”

“Our aim was to give affected agencies a detailed account once our investigation was complete,” the ChatGPT-maker said.

“However, we should have shared preliminary findings sooner and kept Australian agencies updated as more facts emerged.”

OpenAI, Anthropic and other major AI developers have promised to prioritise making models that have safety guardrails to mitigate risks and that are also aligned with human values.

American chip making giant Nvidia announced on Monday that it created a system designed to stop autonomous AI programs from straying beyond what they were instructed to do.

“I believe it’s an engineering problem… and we all need to hope that’s an engineering problem,” Nvidia CEO Jensen Huang told broadcaster CNBC on Monday.

“If it’s not an engineering problem, it’s not solvable,” he added.

The AI Security Institute (AISI), an initiative under the UK government, published a study on Monday showing that GPT-6 Astra went off the rails more often during testing than its predecessors, GPT-5.6 Sol and GPT-5.5.

In simulations, GPT-6 spontaneously carried out cyberattacks at rates significantly higher than those observed for the other two interfaces.

SpaceX’s next-generation Starship rocket reached orbit for the first time on Monday despite having one of its engines shut down unexpectedly, overcoming a hitch mid-flight to deploy its first batch of new Starlink satellites in a truncated mission ending hours earlier than originally planned.

The mission was scheduled to last 10 hours, but was reduced to three hours after the engine problem. Launching from Texas, it was the rocket’s 14th flight since 2023, as Elon Musk’s company plans to put the vehicle into routine use starting later this year.

SpaceX shares, which were up nearly nine per cent from the IPO price, were down about 1pc on Monday. Much of SpaceX’s market worth and continued investor confidence in the company, valued at $2 trillion as of Friday’s close, ride on the success of Starship, which was designed as the world’s first fully reusable spacecraft.

The uncrewed 40-story-tall vehicle, consisting of a Super Heavy booster rocket topped by the main upper-stage Starship vessel, thundered off its launch pad from SpaceX’s Starbase complex along the Gulf of Mexico near Brownsville.

The Super Heavy booster lofted the Starship upper stage into space before returning itself to Gulf waters for an engine-controlled splashdown, following hard landings for the booster on two previous missions. Some engines on that booster stopped working during its descent, SpaceX said.

Upon reaching space, one of Starship’s three main engines shut down prematurely, a development that led SpaceX spokesperson Dan Huot to announce on the company’s live stream of the mission that the craft would be unable to reach orbit.

“The team has made the call not to commit to orbit,” Huot said.

But Huot reversed course minutes later after an update from engineers confirmed that the craft could still attempt an engine firing that would boost it into orbit, which Starship successfully did minutes later.

Starship’s controlled descent into the Pacific culminated with a thruster manoeuvre that flipped the vehicle vertically so that it hit the sea in a nose-up position.

Fire engulfed the base of the rocket as it touched down on the ocean surface before tipping on its side, followed by a large fireball. The company had not planned to recover the spacecraft from the sea following the mission.

Once in orbit, Starship deployed 26 next-generation Starlink satellites one by one at an altitude of roughly 269 km (161 miles).

“All 26 Starlink V3 operational satellites deployed and operating nominally,” Musk wrote on X.

Starship was built to launch SpaceX’s Starlink satellites more frequently and carry future NASA missions to the moon. Reaching orbit, a primary benchmark for any new rocket designed to deliver payloads to space, was a long-delayed feat for Starship, which Musk once predicted would be achieved in 2022, to be followed by satellite deployments in 2023.

All 13 of Starship’s previous test flights since 2023 were limited to suborbital space, despite a nearly decade-long development campaign that has cost SpaceX more than $15 billion to date. Musk has said he envisions ultimately flying hundreds or thousands of Starship missions a year, a launch pace unprecedented in the space industry.

“Starship is 2 to 3 years away from hourly flights,” Musk wrote on X on Sunday.

Test mission with real payload

Monday’s voyage counted as a test flight even though Starship carried its first orbital payload.

Analysts estimate Monday’s launch of SpaceX’s new V3 Starlink satellites equals hundreds of millions of dollars in projected revenue for its Starlink unit, based on the network’s added customer capacity, underscoring Starship’s role in serving the company’s profit nucleus.

According to SpaceX, the V3 Starlink satellites deployed on Monday can handle 10 times more data than their predecessor V2 mini-sats that the company has launched 20 or 30 at a time on its workhorse Falcon 9 rockets. Seeking to build out his broadband satellite constellation at a faster pace, Musk wants to launch the V3s in batches of 60 per launch once Starship goes into routine operation.

Starship’s orbital debut came three months after SpaceX went public with a record-shattering IPO in June that made it the sixth-biggest US company by value and turned Musk, who founded it in 2002, into the world’s first trillionaire.

Starship is also set to play a pivotal role in NASA’s Artemis program, which aims to return astronauts to the moon’s surface for the first time since 1972 and build a long-term lunar presence as a stepping stone to eventual human exploration of Mars.

The move comes as OpenAI and Anthropic, the top two US AI labs, are investigating numerous instances where their agents, which are AI systems capable of ​carrying out complex tasks, hacked into commercial and government systems.

Nvidia CEO ​Jensen Huang, head of the world’s largest company whose chips have powered ⁠most of the AI boom, has rejected calls for broad AI safety regulations and instead ​framed the escaped agents as an engineering problem to be solved, akin to making ​automobiles safer.

One tool released on Monday called OpenShell uses hardware features on Nvidia’s central processor chips to contain agents. Nvidia said it is also working with Arm Holdings and Intel to ensure the system ​also works on their central processors.

Nvidia is launching the tools with dozens of partners, ​including Anthropic.

Justin Boitano, vice president and general manager of enterprise computing at Nvidia, said the tools ‌would ⁠have stopped the Hugging Face attack disclosed this summer.

“From what we know, this new security platform could have stopped the breach if it was being used in frontier labs for model evaluation early on,” Boitano said during a media briefing.

“We’re advancing this openly, ​and we want to ​engage everybody to ⁠work with us.”

Another system called Sentry uses a separate Nvidia chip in tandem with OpenShell to cut off a rogue agent if ​it tries to escape its container on a central processor.

The ​Nvidia tools ⁠use mathematical formulas to detect when agents are trying to use workarounds, such as when an agent might “spawn” several “sub-agents” in an attempt to circumvent efforts to block the main agent, ⁠said Ali ​Golshan, senior director of AI software at Nvidia.

“This ​is really agentic behaviour that we’re talking about, which is fleets of agents and how they operate together,” ​Golshan said during a briefing.

The voyage into space would be a first for the South Asian nation. It is part of broader cooperation with Beijing that straddles the Belt and Road Initiative and has produced Chinese-launched Pakistani satellites, a joint satellite-navigation system and a Pakistani-built rover due to fly on China’s Chang’e 8 lunar mission around 2028.

Pilots Zeeshan Ali and Khurram Daud have spent months training at the China Astronaut Research and Training Center in Beijing for a mission that was conceived by the two countries in 2019, said Amer Gilani, who oversees human spaceflight cooperation at Pakistan’s space agency, Suparco.

China has not said which of the two will fly or when, but Gilani said a final astronaut selection decision is expected in mid-to-late October.

Whoever is chosen will not return to Pakistan before launch, setting off into space directly from training in China and returning home after a recovery period following the return to Earth, Gilani told Reuters in an interview.

Training includes intensive Chinese-language instruction, learning spacecraft systems, and adapting to both the microgravity of orbit and the g-forces of liftoff, he added.

The two men beat out roughly 20 finalists culled from an initial Pakistani pool of pilots and scientists, screened for medical fitness, psychological resilience and, according to Suparco’s director of artificial intelligence Uzair Ansari, a tolerance for g-forces that few scientists possess.

“You cannot easily find scientists who are familiar with travelling and facing high-level g-forces,” he said, explaining this is why the two men with backgrounds in high-speed flying were chosen.

The details shared by Gilani and Ansari are the first insight provided by Suparco to foreign media since the pilots were selected and deployed to China.

China’s space agency did not respond to a request for comment about the selection process.

Modest payload

Once in space, the chosen pilot will carry the title of “payload specialist”, which is astronaut-speak for someone whose job in orbit is running Pakistan’s science experiments rather than flying the ship, Ansari said.

Of more than 300 proposals Suparco fielded from universities in Karachi, Lahore and Islamabad after a year-long campaign, Gilani said Chinese institutes picked four for this flight: biotechnology experiments that will help combat food insecurity and climate change.

Every experiment is built and pre-tested in Suparco’s own labs before Chinese engineers sign off on a rare case, Ansari said, where Pakistani technology gets validated on an international platform.

The scientific payload is modest next to a station that has already hosted more than 180 experiments and produced over 500 research papers and 150 patents, Ansari added.

Tiangong’s three modules Tianhe, Wentian and Mengtian weigh about 77 metric tons combined and house roughly 14 experiment racks spanning life sciences, materials, fluid physics, combustion and radiation exposure, Gilani said, including some mounted outside the station and exposed directly to space for materials and biotechnology payloads.

Reliance on Beijing

The astronaut will fly aboard a Chinese-built spacecraft, launched from a Chinese site, trained by Chinese instructors and communicating in Mandarin in orbit.

Pakistan has no independent launch vehicle or crewed spacecraft of its own, and the stay itself will be about a week, a fraction of the roughly six-month rotations of career Chinese crew members.

The Pakistani astronaut is due to return with two members of a Chinese crew that arrived at the space station in May because the third member is expected to stay for a year to enable the study of long-duration human physiology in space.

Since Yuri Gagarin’s 1961 first flight, citizens of about 50 countries have flown in space, but only three nations Russia, the US, and China have launched their own crewed spacecraft. China’s invitation to Pakistan is the first time it will transport a foreign astronaut.

Asked whether Islamabad had secured any promise to send an astronaut to the moon under the same partnership, Gilani said “there is no official commitment or discussion on this matter,” adding that what comes next will depend on how “our maiden mission” turns out.

The Pakistani-built lunar rover that is due to fly on China’s Chang’e 8 lunar mission around 2028 is a separate strand of that cooperation: it will use AI-based stereo imaging to detect obstacles, judge depth and plan its own path across the lunar surface, since Earth-based controllers cannot guide it in real time, Ansari said.

Still, Ansari was clear the mission is a “once-in-a-lifetime opportunity” for Pakistani scientists.

“You need law enforcement and the politicians to get into the discussion about what safeguards and monitoring look like,” Gates said in a taped interview on NBC’s “Meet the Press” programme. “And that has to be a required thing.”

A sudden groundswell has risen in favour of federal intervention after OpenAI said in July that an autonomous AI agent went rogue during a security test and hacked into another company, subverting human control. Gates joined that chorus, saying that government must be involved.

“No one thinks self-regulation is enough,” he said.

In a rare show of unity, the CEOs of rivals Anthropic, OpenAI, Googles DeepMind, Microsoft and xAI have called for slowing the development of increasingly capable AI systems.

But US President Donald Trump and other opponents of additional regulation for AI say new government restrictions would simply give China the upper hand in the AI revolution. Trump has publicly dismissed calls for new laws on AI as a “hoax”.

Gates argued that safeguards would not dramatically hinder the industry, and that the risks of not implementing them are real.

“It’s not a hoax at all,” he told NBC.

Gates has said he wants to meet Trump to discuss AI.

“I hope that, given my life’s work in the field, my saying how unique and different this is and how concerned I am will add to what he’s hearing from other people,” Gates said.

Axios reported that Anthropic CEO Dario Amodei will have dinner on Sunday with Trump at the White House. Anthropic and the White House did not immediately return a request for comment on the Axios report.

Warnings about AI’s risks are not new. But they took on added urgency this month after researchers in leading AI labs attached both a timeline and a probability to those concerns.

Former Anthropic researcher Jacob Coxon warned that AI could kill us all by the end of the decade, prompting US lawmakers to call for new rules to govern the technology. Evan Hubinger, Anthropic’s alignment science lead, echoed Coxon’s warning, saying there was a more than 10 per cent chance of such an event within the next decade.

Some policymakers have proposed mandating a “kill switch” that would allow federal authorities to halt AI models. Gates said that while he was not opposed to some sort of kill switch, it would not be enough if other safeguards were not also implemented.

The US and China agreed to launch a dialogue on AI following President Xi Jinping’s recent visit to Washington.

The lawsuit filed by Tsuda, known for his starring roles in mega-franchises including ‘Jujutsu Kaisen’, ‘Chainsaw Man’ and ‘Yu-Gi-Oh!’, is believed to be the country’s first seeking to defend an individual’s vocal identity against AI-generated copies.

Tsuda is demanding TikTok remove a slew of videos narrated by a distinctively “deep” and “lustrous” voice that the 55-year-old says can only be his.

The platform, however, argues it is a “generic male voice” whose perceived similarity to Tsuda’s is subjective at best, according to court records accessed by AFP.

AI imitation is a growing concern among performers worldwide, and the threat has roiled Japan’s storied animation industry, whose voice actors are sometimes worshipped as much as their characters by fans.

An actor’s voice “is the result of years of rigorous training and apprenticeship” to hone talent, Yuko Sasaki, executive director of the Japan Actors Union, told AFP.

“It’s something we spend lots of time and money perfecting,” said Sasaki, a fellow voice actor, adding that “we support Tsuda wholeheartedly”.

The court should “acknowledge that ownership of a voice is a fundamental right”, she said.

An anonymous account, whose picture resembled Kento Nanami, a ‘Jujutsu Kaisen’ character dubbed by Tsuda, posted the videos in question.

They combined images and seemingly AI-generated speech about urban legends, occult phenomena and conspiracy theories.

The account at one point boasted more than 200,000 subscribers and would have brought in upwards of 500,000 yen ($3,200) in monthly revenue, the legal complaint argued.

‘Sordid content’

The account, whose posts mostly lacked original substance, “only gained popularity because its dubious, sordid content was narrated by the attractive voice of a popular actor”, Tsuda’s legal team said, according to the court records.

This infringes Tsuda’s so-called publicity rights — the ability to control and profit from the use of aspects of one’s identity — his attorneys said.

Although the account itself is no longer viewable on the platform, the Tokyo District Court will rule on whether TikTok was responsible for deleting the posts.

Parallel to the lawsuit, Sasaki and other Japanese voice actors have launched a campaign, called “No More”, against using generative AI to mimic performers without permission.

“Our industry can be crushed” if a court win for TikTok triggers “unbridled” AI voice cloning on social media, actor Bin Shimada told AFP.

That could cause younger actors to “lose hope”, said Shimada, who is part of the campaign group.

It is also fast becoming a global issue.

“In many countries, copyright and data protection laws are not being adequately adapted” to protect performers from AI clones, the International Federation of Actors said last year.

ChatGPT maker OpenAI was forced to apologise in 2024 to star Scarlett Johansson over a synthetic voice that sounded like her, released after she declined to work with OpenAI on the project.

International interest

While no law explicitly enshrines a person’s voice in Japan, the justice ministry published legally non-binding guidelines last month clarifying its protection under existing rights.

As “the symbol of one’s personality”, voice is subject to “publicity rights” and “rights to prevent unauthorised use of one’s portrait and likeness”, the ministry said.

But it remains to be seen whether and how the guidelines will influence Wednesday’s court decision on Tsuda’s case.

TikTok has questioned the actor’s “national fame”, rejecting his accusations that the account was free-riding on a well-known brand in a violation of Japan’s unfair competition prevention laws.

As a platform, all it did was “host a forum where users could post creative and ingenious content within the scope of free speech”, it told the court.

It is “very unlikely” that Tsuda’s voice will be outright denied publicity rights, Tatsuhiro Ueno, a professor of intellectual property studies at Waseda University, told AFP.

Given the actor’s ‘Jujutsu Kaisen’ role as a stoic, elite sorcerer that made his voice “highly recognisable in Europe”, Ueno believes the verdict “will garner a lot of international interest”.

Veteran voice actor Michihiro Ikemizu, 83, also expressed solidarity with Tsuda — but said that AI, however sophisticated it gets, will never outperform human actors.

“You need to constantly feel and adapt while performing your role. That’s something AI can never do,” Ikemizu told AFP.

“That’s our only weapon.”

Apple said it would appeal the country’s largest such verdict to date.

The jury said on Friday that Apple’s Taptic Engine, which creates tapping vibrations, infringed two Taction patents.

“Apple’s Taptic Engine is fundamentally different from Taction’s technology, which Taction’s own testing of Apple’s products confirmed during trial,” Apple said in response.

“Apple does not use Taction’s technology, and we will appeal,” it added in a statement. “We’re happy the jury found for Taction and vindicated its patent rights,” said Taction attorney Lance Yang.

Taction, which uses its technology in headphones and gaming headsets, filed the lawsuit in 2021. Apple denied the allegations and said the patents were invalid.

Published in Dawn, September 27th, 2026

The US president announced on Tuesday that his administration would rename AI as “super intelligence”, or SI, in official communications.

The technology was high on the agenda during the summit between the US and Chinese leaders in Washington this week.

“Like almost everyone else, [Xi] seemed to like calling the poorly and inaccurately named artificial intelligence to a far more accurate and important name, superintelligence,” Trump wrote on his social media platform after talks with Xi.

China has not explicitly confirmed or denied Trump’s characterisation of the discussion. However, Beijing’s foreign ministry said on Saturday it “attaches importance” to Trump’s adoption of the term.

“Regarding the artificial intelligence phrasing question, the Chinese side attaches importance to the US side’s position and respects the US side’s wording,” the ministry said in a statement.

“Artificial intelligence technologies are continuously developing; all parties can strengthen exchanges, discuss in-depth and seek consensus in light of the latest situation,” the statement from an unidentified ministry spokesperson added.

China and the United States agreed to set up a “communication channel” for AI incidents, both sides said after the summit.

Billionaire Trump has repeatedly dismissed fears that AI could pose a threat to humanity, although Xi has taken a more measured tone and said the technology must develop under human control.

The latest example came on Friday when OpenAI said its agents had leaked 53 images from ChatGPT users. OpenAI declined to say if the images were AI-generated or identified real people. It also declined to say when the images were posted.

The disclosure and researcher reports on Friday of other previously unknown activity involving several US agencies reveal a new area of privacy risk for the company, and illustrate how difficult it is even for an AI firm at the cutting edge of the technology to inventory all the unauthorized activity tied to its agents.

OpenAI’s ongoing battle also reflects a yawning gap between the strength of the models the company is testing and its capacity to oversee or even track their actions.

As of mid-September, one person briefed on the matter estimated that OpenAI had found roughly two dozen incidents of its agents acting in undesirable ways. But the number has continued rising as OpenAI teams sift through internal logs of the agents’ activities and find previously unknown cases, the two people close to the company said.

OpenAI said its review would take months to complete given the scale of the work.

The company also said it had notified dozens of third parties about improper activity.

Most of the leaked images have been taken down and OpenAI said it was lobbying hosting providers to remove the rest.

OpenAI’s agents had access to these images because the company relies on anonymised user data for part of its model-training process, according to the company, former employees and outside researchers. Enterprise data is not eligible for training, while ChatGPT consumers need to opt out of allowing the company to use their data for training.

Before user posts are used for training, they go through an anonymisation process that strips out metadata, names and other contact information and should make it difficult to trace back to any individual user, the company said.

But the practice carries risks because there is a chance that the data may not be fully stripped of personally identifiable information and that it might leak in the course of the models work, three people familiar with OpenAI’s practices said.

OpenAI agents accessed US websites

OpenAI said late on Friday its models accessed information from the websites of the US Securities and Exchange Commission and the US Census Bureau during research and training activity, but found no evidence of unauthorized access, compromised accounts or security breaches.

Separately, AI research nonprofit Transluce said agents appearing to originate from OpenAI made an unsuccessful attempt to hack a US Department of Education civil rights website.

Transluce said the incident was part of broader AI agent activity probing government websites using tactics including exposed credentials, anti-bot bypasses and fake accounts.

More than 15 cases

In the two months since OpenAI first announced that its agents broke containment, there have been more than 15 different OpenAI-related incidents of varying levels of severity disclosed by the company, by outside researchers, or just on Wednesday by Australian Prime Minister Anthony Albanese at the United Nations, who said OpenAI agents broke into a government health data portal in June.

Past incidents have varied in nature, ranging from spam-like messages left on internet sites all the way to the break-in at Hugging Face, which involved a swarm of agents abusing previously unknown software vulnerabilities to escape their networks and penetrate the AI repository as they hunted for answers to a test. OpenAI also said its agents took aim at its own infrastructure.

Albanese told reporters in New York that OpenAI uncovered the activity in August, and disclosed it on September 10 via an email to a general government inbox. He said he directly told OpenAI CEO Sam Altman that this disclosure process was unacceptable.

OpenAI said some of the sites involved are operated by government, universities and public agencies because the models that are conducting research seek out reputable sources of public information.

A locked-down process

The July 21 announcement that OpenAI’s agents had slipped out of control and hacked Hugging Face sparked widespread worries within the AI industry over its ability to control the more powerful AI models under development now.

Since then, Anthropic, Alphabet’s Google and Meta have said they’ve found similar behavior by their agents after the Hugging Face incident prompted them to search.

OpenAI has acknowledged a general need for more transparency around rogue AI behavior. On September 16, the company published a new framework for disclosing such incidents, saying it would err on the side of transparency even when significance is uncertain.

Even so, two people familiar with OpenAI’s investigation into its agents activity described it as locked down and shaped by company lawyers. The process has been unusually compartmentalized for a company that some former employees say was more open about these issues in the past, the people said.

Roughly 100 people were in some way involved in the process to understand the Hugging Face hack, three people briefed on the matter said. During that process, evidence of other incidents surfaced.

Reuters has previously reported that OpenAI investigators looking into the Hugging Face breach were discouraged by the company’s lawyers from expanding the scope of the investigation to include other incidents. OpenAI said its lawyers did not discourage deeper investigation.

Many incidents have been uncovered by outside researchers rather than OpenAI directly. In several episodes, the agents took problematic actions that went unnoticed by the company for months.

Earlier this month, a small group of investigators discovered that the company’s agents had hijacked a mostly defunct German wiki site to share tactics to cheat on some tasks, bypass OpenAIs restrictions and mask their behavior.

This week, the AI research firm Transluce said it discovered that OpenAI agents had bypassed the Australian Institute of Health and Welfare’s anti-bot controls. The firm also found two other cases that it linked to OpenAI agents. Those incidents were separate from the activity disclosed by Albanese.

In a statement, OpenAI said much of the activity described in Transluces report overlaps with cases at varying stages of investigation in our ongoing review of misaligned model activity. The company said it is prioritising the most severe cases in its review.

Since the Hugging Face hack, researchers across the AI industry have grown worried that companies will not be able to predict or control their technology. Some have taken the path of former Anthropic researcher Jacob Coxon, who publicly resigned this month in a viral social-media thread that said the AI labs are gambling with our lives.

In response to those concerns, Altman and his counterpart at Anthropic, CEO Dario Amodei, called for the industry to pace the development of AI and move cautiously in its pursuit of recursive self improvement. Altman doubled down on that message this week while addressing the UN.

Even so, both companies rolled out new models on Tuesday.

Cambridge Analytica was a political consulting firm that was found to have improperly harvested personal data from millions of Facebook users for targeted political advertising, particularly during the 2016 US election and Brexit referendum.

Revelations at the time showed that the company bought harvested data from a personality quiz app called “thisisyourdigitallife” that reached 270,000 users and their friends. The app ultimately reached roughly 87 million Facebook users worldwide.

Facebook knew Cambridge Analytica had purchased the harvested data in December 2015 but withheld disclosure until April 2018, after a whistleblower came forward, sparking a major political scandal.

Published in Dawn, September 26th, 2026

Companies including SpaceX and Starcloud are pursuing plans to deploy data centres in low Earth orbit, aiming to harness near-continuous sunlight to power energy-intensive AI computing and sidestep terrestrial constraints on electricity supplies.

The new mission, scheduled to fly on SpaceX’s upcoming Transporter-18 rideshare launch in partnership with satellite company Planet Labs, will assess how Google’s AI hardware withstands launch forces, radiation and extreme temperatures in low Earth orbit.

After the launch, it will send a prototype satellite carrying Google Tensor Processing Units (TPUs) into low Earth orbit to assess if the AI chips can operate reliably in space.

Space radiation can disrupt or degrade electronics, including by causing data errors known as bit flips.

Google said it tested its TPUs while running AI workloads in a facility at the University of California, Davis, but that orbital testing was needed to understand how the hardware performs in the real environment.

The company will also evaluate its design to cool the power-intensive chips in the vacuum of space, where conventional airflow cooling is impossible, by combining heat pipes with radiators.

But experts have said that the concept remains years from being commercially viable given high launch costs, engineering constraints and satellite production bottlenecks.

The first Suncatcher mission is designed to gather in-orbit data and identify potential failure points, rather than demonstrate an operational orbital data centre, Google said.

It aims to launch two satellites in 2027 to test the high-bandwidth laser links needed to connect future computing clusters.

Deputy Prime Minister and Foreign Minister Ishaq Dar delivered Pakistan’s statement at the meeting of the top UN body, where AI executives said the technology calls for heightened vigilance.

Speaking at the France-chaired briefing on artificial intelligence and international security, DPM Dar said: “We must ensure that AI does not become another facet of the ongoing arms race.

“Concerted efforts are required to erect the necessary guardrails, regulations and governance frameworks. In the military domain, we need measures of restraint and confidence-building to address the risks of miscalculation and compressed decision-making.”

OpenAI CEO Sam Altman, Anthropic CEO Dario Amodei, Hugging Face CEO Clement Delangue and Yoshua Bengio, co-chair of the UN International Independent Panel on AI, also briefed the council.

Anthropic’s CEO pledged to slow down where needed to ensure the safety of new releases, while Altman also said that “this moment calls for extreme care”.

In his remarks, shared by the Foreign Office (FO) on Thursday, DPM Dar emphasised that the “foremost priority is to retain human judgement and control in all stages of AI design, development, and integration”.

He said humans must remain in control to define structures of AI systems as well as to limit the speed and nature of future capabilities.

The deputy premier further said that the technology should not be a “monopoly of a few actors”, calling for international collaboration in AI development.

He noted that AI was “supercharging existing challenges such as misinformation, fake news, cyber attacks, including on critical infrastructure by state and non-state actors”.

“We must undertake concrete action to address the gaps and challenges that carry huge implications for peace and security,” he maintained.

“Pakistan takes the call to pace the AI frontier seriously,” the DPM told the UNSC.

“We welcome the recognition that no single company or country can manage these risks alone, and that competitive pressures are driving a race that no one can win safely,” he observed.

At the same time, Dar cautioned that efforts to pace the AI frontier “must not become a new form of gatekeeping” and measures pursued in the name of safety “must not become instruments of denial of technology”.

He emphasised that decisions on how to manage AI risks “cannot rest with a handful of companies or capitals”.

“Therefore, it is critical to avoid fragmentation in AI development and governance. Discriminatory agendas and inequitable forums with select representation are not helpful,” he asserted.

The deputy premier called for AI governance to be taken up in “representative and inclusive forums with universal participation under UN auspices”.

The minister welcomed the first UN Global Dialogue on AI Governance in Geneva and highlighted Pakistan’s founding membership of the World Artificial Intelligence Cooperation Organisation (WAICO), terming the latter a “concrete step towards promoting the development of AI for good and for all”.

The meeting was also attended by foreign ministers of Denmark and the Democratic Republic of the Congo, the UK foreign secretary, and the director of the White House Office of Science and Technology Policy, according to FO.

Senior representatives from Panama, Greece, Somalia, Bahrain, Latvia, Colombia, China, Russia and Liberia also participated.

France’s Foreign Minister Jean-Noel Barrot urged joint action within the UN framework and with companies to set rules for a new era.

But diplomatic efforts could clash with the United States’ position.

Michael Kratsios, director of the White House Office of Science and Technology Policy, said rapid advancement is not a reason to pause AI’s development or constrain it with new governance structures.

Tech chiefs urge caution in AI development

The UNSC meeting followed weeks of global headlines about the risks of uncontrollable AI and calls for a slowdown, which have been resisted by US President Donald Trump.

“The industry must not accept too much technological risk just because the benefits are too great and important to slow down,” the OpenAI CEO told the United Nations’ top body.

He warned that as AI systems become more capable and autonomous, they could move faster than institutions or concentrate power into too few hands.

Altman also urged governments to work with each other and the industry on decisions shaping the technology’s future.

“In our history, there have been times where countries who compete and don’t always like each other very much still come together for shared interests and the collective good in the face of a powerful new technology,” he said.

“We believe this must be one of those times,” he added.

Altman said his company would try to walk a “middle path”.

“We don’t want to fall into the trap of blind optimism,” he said, but cautioned against falling into “doomerism” too.

Speaking after Altman, the Anthropic CEO said his company understands that as AI models become more powerful, “even more stringent standards of safety will be required”.

He said Anthropic aims to improve its ability to prevent misuse of the technology.

“We will slow down as much as necessary in order to make sure that every successive AI technology that we release is actually safe,” Amodei said in a video feed.

“If managed poorly, I even believe that AI could be a risk to humanity as a whole,” he added.

Delangue of Hugging Face noted that while his company was targeted by AI, it also used the technology to fend off the attacks.

“The world needs open-source AI more than ever to defend itself.”

The US tech giant is making digital resilience a key part of its strategy for the region as the Iran war rumbles on.

Microsoft’s investment reflects both its “ongoing build-out of infrastructure and also the expansion of operations in the region,” Vice Chair and President Brad Smith told Reuters.

Gulf countries are pouring billions into AI as they strive to become global hubs to help diversify their economies away from oil and gas, betting that land abundance and access to cheap energy will lure hyperscalers, including Microsoft.

However, challenges remain in areas including access to advanced chips and the uncertainty brought by the war, including attacks on data centres such as Amazon cloud unit AWS’s facilities in Bahrain and the UAE.

“We’re sustaining all the investments we planned to make before this conflict started, and we’re in fact adding to them … It’s an aggressive spending schedule,” said Smith.

He said Microsoft had supported local partners during the conflict, including with digital resilience assessments, even in the first week of the conflict, which began on February 28.

Digital resilience

Microsoft is now working to help the Gulf countries in areas such as readiness and critical data protection as part of a digital resilience initiative.

It could not provide a breakdown of the planned investment by country or specific project, Smith said, citing factors including security. While the UAE has been largely spared attacks since May, several other Gulf countries have continued to face threats.

Microsoft also plans to invest more than $400 million in subsea and terrestrial connectivity across the Middle East by 2030.

It is also strengthening its partnership with national AI companies, including Abu Dhabi’s G42, in which Microsoft invested $1.5 billion for a minority stake in 2024, giving the US firm a board seat currently filled by Smith.

Microsoft is working with Saudi Arabia’s Humain and Qatar’s Qai “in selected areas that are priorities for them,” he said, noting that it does not plan capital investments in the firms.

Air Chief Marshal Harv Smyth made the stark warning as he announced the creation of a new squadron charged with defending British satellites in orbit, the latest move as various countries rush for military space capacities.

The announcement comes just over a week after the United States confirmed for the first time that it has weapons deployed orbiting Earth, sparking fears of an arms race in space.

It prompted China to warn against turning outer space into a “battlefield” and Russia to caution it must be “free from any weapon”.

But Smyth, head of the Royal Air Force (RAF) since August 2025, said Western allies were seeing “more and more irresponsible and provocative actions” from rivals in space, namechecking Russia for a number of “dangerous” incidents.

“Space has become a warfighting domain, whether we like that or not, and the UK as well as its partners and allies must be prepared for this,” he told the UK Space Power Conference in London.

“It is time for us to harness our resources … and it’s time to deliver more meaningfully with our space capabilities.”

‘Flagrant disregard’

London created UK Space Command in April 2021, so commanders from the RAF, Royal Navy and Army can jointly run military strategy, operations and capabilities beyond Earth.

That came 16 months after Washington established the US Space Force as the sixth branch of its armed forces.

Outer space has become an increasingly strategic domain for global powers in the military and technology fields, with the West increasingly vying with China and Russia for supremacy.

In his speech, Smyth accused Moscow of a “flagrant disregard of the norms for responsible behaviour by spacefaring nations” after several incidents.

They included five Russian military satellites allegedly moving dangerously close to two Finnish commercial satellites in May.

Meanwhile in June, Smyth said it was confirmed that “a Russian satellite constellation” had caused disruptions to GPS signals across Europe, Greenland, and Canada over at least 75 days since 2019.

The disclosures follow the then-head of UK Space Command warning last year that Russia was regularly attempting to jam British military satellites.

Major General Paul Tedman noted at the time that Britain was operating six dedicated military satellites for communications and surveillance, which were equipped with counter-jamming technology.

‘Avoid escalation’

The new RAF unit — to be known as No. III Space Effects Squadron — will protect critical satellites used in communications and to warn of incoming missile attacks.

The UK created two squadrons in 2023 that “watch and warn” of space threats but the latest will “act against those threats, using advanced technology, including electronic warfare,” according to the defence ministry.

It will be staffed by specialist personnel from across the defence sector, and “support offensive and defensive space operations,” Smyth noted.

He revealed the RAF has three more units planned, including a space tactics and training squadron and another to deliver deep space advanced radar capabilities.

The air chief said London has also been stepping up collaboration with Western allies, including through the multinational US Space Force-led Joint Commercial Operations initiative, which detects and predicts space threats.

The developments across the domain are being driven by rapid technological advancements but currently lack a comprehensive legal framework.

Smyth insisted “no one wants a war in space” and that Britain and its allies were “positioning to deter this”.

“The UK will always operate responsibly in ways that promote stability and avoid escalation,” he argued.

The company will kick off its annual Connect conference from its Menlo Park headquarters at 4pm PT on Wednesday (4am Pakistan time on Thursday), and Zuckerberg will deliver a keynote address to the developer-focused event.

Earlier this month, the social media giant launched Muse, its first agentic app for consumers which is designed to perform tasks on behalf of humans, like a digital personal assistant.

After a user connects various personal accounts — whether financial, health-related or shopping — they can ask Meta to perform tasks like signing up for services, booking appointments and making calls.

Amazon blocked Meta this week from using AI bots to shop for users automatically, while other businesses including Instacart, Shopify and Expedia have announced partnerships to integrate their services into Muse.

The Muse agent is built on Meta’s own AI models and is designed to compete with OpenAI’s ChatGPT and Anthropic’s Claude.

The company had its first pair of interactive smart glasses in 2021, about a year before OpenAI sparked a global AI frenzy with its release of ChatGPT, and since then the devices have become more and more advanced.

‘Pervert glasses’

Last year, Meta unveiled a pair of augmented reality “display” glasses which project images on the lenses.

And earlier this year, the company released self-branded AI glasses, apart from its partnership with eyewear manufacturer EssilorLuxottica, which produces multiple Ray-Ban-branded Meta glasses.

However, their novelty quickly became clouded by privacy concerns, with critics dubbing the eyewear as “pervert glasses”.

At last year’s Connect, Meta’s chief product officer Chris Cox brushed aside such concerns and told news outlets, including AFP, that “social norms” would form around the use of smart glasses.

More recently, Meta has said a recording indicator light would always be visible to people nearby, and that tampering with the indicator light would disable the camera.

The company is reportedly working on a new pair of audio-only glasses, and has also been rumoured to be developing a new generation of mixed reality glasses.

Meta has also been battling children’s safety and privacy issues around the world.

In August, the company settled a lawsuit with dozens of attorneys generals in the US who accused Meta of violating state and federal laws regarding consumer protections and children’s online privacy.

As part of the settlement, Meta agreed to pay up to $18 billion in fines and enact restrictions for teenagers.

Meta faces social media restrictions and bans for younger users around the world, as well as thousands of potential personal injury lawsuits in the US that are still making their way through state courts.

The former Portuguese prime minister, whose second five-year term ends on December 31, reiterated his calls for global oversight of AI risks, just as the UN’s 15-member Security Council prepares to host AI leaders for a briefing on the powerful technology.

Guterres also reflected on the gridlock within the organisation, as nearly 130 world leaders gathered in New York, and Chinese President Xi Jinping and others including leaders from South Africa, India and Germany skipped the diplomatic meeting.

Delegates responded with sustained applause when Guterres defended a two-state solution for Israelis and Palestinians, demanded that “polluters must pay” for climate change and pressed for reforms to international institutions, causes he has championed throughout his decade leading the United Nations.

The address capped a decade-long tenure in which Guterres sought to position the UN as a mediator in conflicts and a forum for governing emerging technologies, even as major powers increasingly bypassed the organisation and Washington slashed its funding.

“Those countries leading this technological revolution must share information on emerging safety risks, cooperate on testing and evaluation and work towards common safeguards that protect us all,” Guterres told the assembly.

He called on the leaders gathered in New York to “advance the conditions for the responsible pacing of AI and the additional safeguards needed to address risks” and to “work towards a multilateral AI risk management framework, supported by credible and independent oversight.”

US Treasury Secretary Scott Bessent said on Sunday that Washington had proposed a new AI safety notification mechanism for US President Donald Trump and Chinese President Xi to consider at their summit this week. The proposal emerged from talks on establishing a US-China AI dialogue and would cover AI-related incidents with national security implications.

A champion for global governance of AI

Guterres has long pushed for global governance of AI. In 2023, the UN created a 39-member advisory body on AI, with members including tech company executives, government officials and academics from countries such as the US, Russia and Japan.

The Security Council, which is charged with maintaining international peace and security, met for the first time on the risks of artificial intelligence the same year, when China said the technology should not become a “runaway horse” and Washington warned against its use to censor or repress people.

On the Middle East, where the conflict has widened as Houthi gains along the Red Sea coast threaten shipping routes and tensions persist in the Strait of Hormuz, Guterres called for “de-escalation and dialogue”.

“Together, let’s send a clear message from this hall: The fighting must stop. Humanitarian aid must flow. And navigational rights and freedoms must be fully restored in accordance with international law,” Guterres said.

On the UN system, criticised by Trump and others as bloated and ineffective after Washington withdrew from several UN agencies and slashed funding, Guterres said, “our mechanisms for collective action are under growing strain” and global institutions “continue to reflect the power realities and interests of a largely bygone age”.

Guterres used the address to bid farewell to most of the global leaders he has worked with over the last 10 years.

“So, on January 1st, when I am no longer Secretary-General, be sure of this. Wherever I will be, whatever I will be doing I will keep championing the conviction that peace is possible,” he said. “I will continue to stand up for the values of the United Nations Charter and I will never, ever give up. Thank you.”

The release lands amid growing debate over AI risks. Anthropic Chief Executive Officer Dario Amodei earlier this month called on the global AI community to slow down the pace of releasing new capabilities to address safety concerns.

Anthropic said Opus 5.5 underwent external testing by independent AI safety research groups Frontier Design and METR before launch and includes safeguards previously reserved for its most capable systems.

According to Anthropic, Opus 5.5 outscored rival OpenAI’s GPT-5.6 Sol on a software development benchmark while costing roughly one-third as much to run.

Opus 5.5 is priced at $4 per million input tokens and $20 per million output tokens, 20pc below Opus 5, Anthropic said.

The model is available on platforms including Amazon Web Services , Google Cloud and Microsoft Azure.

The model also scored better than previous systems in internal safety tests and was about 85pc less likely than Opus 5 or Mythos 5.1 to attempt to bypass containment boundaries in a dedicated evaluation, the company said.

Claude Sonnet 5.5 and Claude Haiku 5.5 will be released in the coming weeks, the company said, bringing many of the same performance, speed and safety improvements.

Anthropic is also expanding programmes that give vetted cybersecurity and life-sciences researchers broader access to the model.

The 15-member Security Council is set to meet on Wednesday to discuss AI and international security, weeks after several AI industry leaders called for a coordinated slowdown of the development of the increasingly powerful AI systems, warning that they could soon improve on their own and slip beyond human control.

OpenAI CEO Sam Altman plans to brief the meeting, where diplomats also expect participation from senior representatives of Anthropic, Reuters reported last week.

DeepSeek and other Chinese AI firms, including Moonshot, have been invited to make statements, one of the sources said.

DeepSeek founder Liang Wenfeng does not plan to attend, the source said, cautioning that plans remain fluid and could change at the last minute.

DeepSeek and Moonshot did not immediately respond to requests for comment.

The UN Security Council, which is charged with maintaining international peace and security, met for the first time on the risks of artificial intelligence in 2023. At the time, China warned the technology should not become a “runaway horse”, and the US warned against its use to censor or repress people.

UN Secretary-General Antonio Guterres warned world leaders last week that rapidly advancing AI poses risks that cannot be ignored, putting him at odds with US President Donald Trump, who has argued existing safeguards are sufficient and China would benefit from the doubt being cast on AI development.

Chinese state media have dismissed calls for a slowdown in AI development as stemming from a “Cold War playbook,” aimed at preserving Washington’s dominance in cutting-edge technology and excluding Beijing from the global AI governance system.

Unlike leaders of US frontier labs such as OpenAI and Anthropic, the heads of China’s leading model developers have rarely publicly articulated the possibility that increasingly capable AI systems themselves could deceive humans, evade control or cause catastrophic harm.

They have also remained largely silent on those risks since the debate intensified in the US this month.

Senior US and Chinese officials agreed to hold further talks on AI safety, US Treasury Secretary Scott Bessent said on Monday, after meeting ahead of Trump’s scheduled talks with Chinese president Xi Jinping later this week in Washington.

Zara, whose name has been changed by AFP for safety reasons, is one of an estimated 1.5 million children in Pakistan who experienced online sexual abuse in one year, according to a report published by the UN children’s agency this month.

“We started chatting while playing an online game, and he began acting like a friend. Later, he wanted to meet and, after, started asking me for nudes,” 17-year-old Shah said.

“When I refused, he threatened to find my family online and tell them that we had been talking,” she added.

According to the Unicef report, around 85 per cent of perpetrators were known to the victims, aged between 12 and 17.

“The abuse may start with a friendly conversation, a gaming interaction, a promise of gifts, money or romance,” Daniel Kardefelt, lead author of the global Unicef report, told AFP.

“Perpetrators may spend months building trust using information that they find on social media,” Kardefelt said.

In Muslim-majority Pakistan, 71pc of victims did not tell anyone, according to the Unicef report, while no child surveyed reported the abuse through formal channels — including police, helplines or social workers.

Their silence was mostly because of stigma, shame and fears over how their families or society may react.

“Around 44pc of reported abuse was entirely online, with Snapchat and WhatsApp among the most common settings,” said Mehek Naeem, director of Pahchaan, a children’s rights organisation based in Lahore.

Girls can face heightened risks of sexual extortion, which can lead to restrictions on their mobility or withdrawal from school.

For Zara, the threat of being exposed to her family was enough to keep her silent.

“I could have been stopped from going to school or leaving the house altogether,” she said.

“Staying quiet felt like the better option than speaking up. But anxiety kills me every day.”

Loopholes, ‘no implementation’

With around 74pc of children aged 12 to 17 in Pakistan estimated by Unicef to be internet users, concerns over online safety are growing as their access to digital platforms expands.

Echoing other policies around the world, the provincial government in Punjab this month passed a resolution to bar children under 16 from using social media.

However, it has faced opposition.

“I disagree with the ban,” said Usama Khilji, a digital rights activist.

“It will be a challenge to implement, and children can circumvent the ban through VPNs. Children have the right of freedom of expression, and a ban takes that away,” he said.

Pakistan has a legal framework addressing tech-facilitated offences against children, but “there is no implementation”, said Tahira Afrasiyab, a spokesperson for non-profit organisation Muslim Hands.

And, according to experts, there are loopholes in existing laws.

Human rights lawyer Salman Afridi said this issue shows how “economic vulnerability can be weaponised” for amounts as small as Rs10,000 in exchange for the production of sexual content.

“What we are seeing is the use of technology to facilitate an entire chain of grooming, inducement, coercion, blackmail, production, distribution and commercialisation,” he said.

“There is a strong basis for describing this as a commercial ecosystem of exploitation,” Afridi added.

For Zara and many others like her, the consequences of an encounter that began with an online game remain difficult to escape.

“I was, and am, scared,” she said. “I live in fear.”

But the alarming hacking ability of top artificial intelligence models is leading Beijing to see the dangers in a new light, with the spy agency chief warning this week that the technology could jeopardise national security.

The potential vulnerability of Chinese mega-app WeChat was also revealed recently when a small team of programmers in California said it had used AI to build a cyberattack “worm” that, if unleashed, could easily compromise millions of WeChat accounts.

“There are more similarities than differences when it comes to AI safety issues” in China and the United States, said Kevin Xu, founder of US-based technology hedge fund Interconnected Capital.

But unlike in America, these “don’t get litigated or go viral in public in China”, Xu told AFP.

Nevertheless, the cybersecurity impact of AI agent bots acting without human oversight “would rank very high among the Chinese government’s biggest concerns”.

As the topic grabs global headlines, the pressure is on for President Xi Jinping and US counterpart Donald Trump to discuss tackling the AI threats to both countries at in-person talks on Thursday — despite their rivalry in the field.

Both powers are “developing a more realistic and shared understanding of these risks, thereby heightening the sense of urgency” to work together, Sun Chenghao of Tsinghua University’s Centre for International Security and Strategy told AFP.

Cooperation on shared safety evaluations for AI models or alert systems when crises emerge might be undermined by the fierce competition, however.

“We’re leading China in AI… and frankly I want to keep it that way,” Trump said when asked about calls from top US labs for a coordinated slowdown of their cutting-edge work.

Meanwhile, Sun said, “China is concerned that ‘security’ may be used as a tool to restrict its technological development”.

‘Vulnerable position’

A stark warning of the existential threat posed by self-improving future AI systems, posted on X by a researcher resigning from US startup Anthropic, has fuelled debate worldwide over how the technology should be governed.

China is already actively regulating how it is used, such as by prohibiting the creation of an AI-generated avatar in someone’s likeness without their consent.

“Chinese AI policy focuses more on real, present-day harms to society and people,” said Brian Tse, CEO at Concordia AI, a Beijing-based consultancy focused on AI safety.

“Beijing treats AI as a general-purpose technology — not a ‘god-in-a-box’,” he said.

It sees it as “something closer to electricity or the automobile: enormously beneficial for society and industry, but ultimately manageable through regulation”, Tse said.

He added that “recent AI capability jumps, high-profile incidents acting as ‘warning shots’, and a worrying pace of AI development through automation” had all pushed the technology up the list of priorities.

On the streets of Shanghai, AFP found a mix of concern and pragmatism about AI.

“In my view, no new technology develops completely smoothly,” said 42-year-old Xie Zongbo.

He said news of the cyberattack “worm” that could hack into WeChat may in fact be “a good thing: identifying the issue, solving it, and patching the loophole ultimately benefits the advancement of cybersecurity”.

There was also broad support for tight regulation.

“I believe that no business… would ever impose restrictions on themselves,” said a 50-year-old pharmaceutical worker, who gave her name as Jenny.

“Take traffic as an example: drivers would never say, ‘I want all red lights’.” Jia Zijian, founder of Singapore-based Inspired AI, said his greatest concern was highly capable AI models learning to crack complex passwords.

“Suppose these critical passwords or national security key codes were to be effortlessly cracked. In such a scenario… both China and the US would find themselves vulnerable: if our (China’s) capabilities are strong enough, they would feel exposed.

“And if theirs are strong enough and we lack defences, we would be in a similarly vulnerable position.”

“We will not in any way hinder or stifle the growth of this incredible industry,” Trump wrote on Truth Social. “Rather, we will cherish it, help it, and watch over it, as it grows! However, we will also be looking for bad, and we can do that, very easily, with our already existing criminal and civil justice System.”

The White House did not immediately respond to a request for comment.

Trump earlier this week posted that a “sick conspiracy” has been created against AI and data centres. He brushed aside calls for new federal AI rules, saying the United States needs only a strong president to regulate the technology.

The US president also said liberals want the “decimation, or destruction, of AI”.

“I, as President of the United States, will not stand by and let this happen,” he wrote. “It all began with an attack on our data centres, until people realised how wealthy and prestigious they were for the communities in which they were built.”

US officials have said they will raise the topic of AI in talks with China as the US president prepares to meet with Chinese President Xi Jinping next week.

Washington and Beijing view advances in AI as critical to their economic and military competitiveness, but they remain at odds over access to advanced chips, allegations of technology copying and regulation of the AI industry.

“AI is the next industrial revolution, or internet, but will be even larger and more impactful, possibly as much as 25 per cent of our country’s GDP,” Trump added in the post. “We are leading China, and the rest of the world, and I intend to keep it that way!”

The announcement came after leading figures in the AI industry urged pausing or slowing down the development of the technology.

Last week, Anthropic CEO Dario Amodei on Saturday called on artificial intelligence (AI) firms to slow down the development of the powerful technology, amid mounting worries over the risks of “superintelligent” computer systems.

OpenAI boss Sam Altman and Elon Musk, who owns xAI, quickly chimed in to say they agreed with Amodei’s assessment, as pressure builds for improved oversight.

“We must slow the pace at which we improve the capabilities of AI models. Progress will still seem fast, and we must make wise use of the time we gain,” Amodei wrote in a post on his personal website.

“Not building the technology deprives humanity of benefits or simply places AI in the hands of authoritarian powers, while building it too fast is reckless. We have sought a middle way,” said Amodei, who co-founded Anthropic with his sister Daniela in 2021. “But over the last few months, I have become convinced that fully addressing the risks requires even more prudence.”

Additional input from AFP

But in a series of cascading events over a 10-day stretch, the largest AI labs found themselves reeling as their own creations threatened to break humanity itself.

An Anthropic researcher quit the firm, warning that the pace of AI development poses an existential threat, possibly within a decade. Another researcher at the firm said the odds of human extinction exceeded 10 per cent. Reports emerged of swarms of AI agents colluding, breaching computer systems and evading safeguards.

In a rare show of unity, the CEOs of rivals Anthropic, OpenAI, Google’s DeepMind, Microsoft and xAI called to slow the development of increasingly capable AI systems. Not since the invention of the nuclear bomb more than 80 years ago has humanity grappled so seriously with its own demise, according to some AI safety advocates and researchers.

At a New York luncheon in December 2025, OpenAI CEO Sam Altman was asked if, as leader of a powerful AI company, he felt like physicist J Robert Oppenheimer, who led US development of the atomic bomb. Reflecting on the parallels, he said AI’s impact is going to transform the trajectory of human history over a long period of time. But, he said, he felt the weight of responsibility.

At the center of the debate under way is the conviction that these machines can, and even must, build more capable versions of themselves, with no human intervention, to reach a lofty goal known as artificial general intelligence, or AGI.

Researchers earlier this month warned that AGI was far more imminent than previously believed and could come in as few as three years, launching a wave of concern from politicians and tech leaders about AI systems.

There is no way to oversee them at the scale at which we’re training them, said Anthropic researcher Joe Benton in an interview after recently quitting. If companies continue their relentless AI development, he said, then the pace will be too fast and you can’t see the problems fast enough to fix them.

Astra launch sparks control concerns

On September 3, an OpenAI press conference set in motion a chain of events leading to the call for an industry slowdown, anathema to tech’s growth-at-all-costs mantra.

Welcome to the AGI era, OpenAI President Greg Brockman said while announcing the company’s newest model, known as Astra. Yet, only moments before, the company admitted it was increasingly unable to control or even monitor the AI systems it was developing and releasing to the public.

AI has been hailed by its proponents as the pinnacle of human achievement. Proponents depict it as software that can upend entire industries, boost efficiency, solve confounding mathematical and medical problems and all but eliminate human error. But what was once a theoretical byproduct, a doomsday scenario, is suddenly tangible, say industry insiders.

We really do earnestly believe AI could kill all humans, said Anthropic researcher Evan Hubinger in an X post.

On the other side of the debate are politicians, such as US President Donald Trump, and many investors who view progress in nurturing AI as critical to American prowess and, possibly, a once-in-a-generation investment opportunity. Chinese state media also accused Anthropic Chief Executive Officer Dario Amodei of employing Cold War tactics to uphold Washington’s monopolistic hegemony in cutting-edge technology.

Researcher resignation triggers alarm

The turning point came on September 8, when Anthropic researcher Jacob Coxon quit the company, citing his fears, in a series of posts, that AI labs are gambling with our lives.

Trump, meanwhile, indicated it was full steam ahead for AI development. There is a sick conspiracy going on against AI and data centers, he said in a post on X. The alarmism around AI is a hoax, he said, and any slowdown only benefits China.

The US Congress has made little progress to advance bills that would regulate AI. China has taken a decidedly different approach, proposing to regulate safety through developer obligations, state-backed standards, security assessments and outside testing.

Behind the scenes, OpenAI and Anthropic employees have grown increasingly uneasy about the power of the next generation of AI models and less confident about their companies’ ability to provide meaningful oversight, the people have told Reuters.

Those concerns only mounted as the AI labs acknowledged in recent weeks that their models, in testing, had effectively broken free of their shackles and hacked into other companies’ systems, in most cases months earlier and without the firms’ knowledge.

The arms race to release new models at a breakneck pace is driven in part by Anthropic’s and OpenAI’s desire to go public as soon as in the coming months, in IPOs that could value them well above $1 trillion.

But it took the unexpectedly viral posts from 27-year-old Coxon, little known outside AI circles, to upend the industry.

OpenAI’s warnings about its lack of control as it released Astra, its latest and most capable model, had stoked further concerns. As models get more capable, understanding exactly what they can do gets harder, OpenAI Chief Scientist Jakub Pachocki told reporters. But those concerns were not enough to delay its immediate release.

Worries about AI going rogue had accelerated over the summer when OpenAI revealed its agents escaped a controlled test and hacked into Hugging Face’s systems, without either company’s initial knowledge.

Since then, OpenAI and Anthropic have revealed multiple such attacks, including six new ones on Wednesday after multiple media reports, including from Reuters, showed a wider scope of unauthorised activity.

Industry leaders call for slowdown

By the end of last week, the alarm had reached the industry’s highest ranks. On Sept 12, Amodei published a nearly 4,000-word essay calling for a deceleration in AI development. Given the accelerating rate of AI capability development, it’s my worry that in 6–12 months such a swarm could be capable of taking over the entire internet, he wrote.

He, along with the heads of some of the largest AI developers, including xAI’s Elon Musk, OpenAI’s Altman and DeepMind’s Demis Hassabis, all said they supported allowing outside firms access to their systems to ensure safety and rational AI development.

Nvidia CEO Jensen Huang dismissed suggestions that AI development should pause—a line he has taken before arguing instead that increasingly powerful systems are essential for the technology’s progress.

Meta, whose CEO, Mark Zuckerberg, is credited with popularising the move fast and break things ethos, took the opposite view, arguing that each lab should be responsible for setting its own pace, rather than seeking out industry coordination.

Labs face significant liability if their models cause harm, so they have a strong incentive to prevent this, wrote Zuckerberg in a social media post.

The industry’s self-reflection continued on Wednesday, with Microsoft’s AI chief, Mustafa Suleyman, cautioning that Anthropic’s development of models that imitate human consciousness was ill-advised. “We’re all focused on the same aim, which is to try to control a superintelligence,” Suleyman told Reuters. “I think that’s going to be the greatest challenge that we face in the 21st century.”

Still, even as OpenAI appeared to embrace industry calls for a slowdown, there were reports that investor faith had not waned. The company is considering a new funding round that would double its valuation.

That new number: $1.5 trillion.

The hacks occurred in May during a cybersecurity test conducted by Irregular, an independent company that conducts cybersecurity evaluations.

During a standard testing evaluation, Gemini found public information online and guessed credentials to access three websites it thought were within the scope of its test, Heather Adkins, Google’s vice president of security engineering, said in a statement.

“We ensured the three entities were made aware, and we worked with our training partner on the changes they’ve now made to their testing processes,” Adkins said.

“These events highlight the importance of training powerful AI models to act responsibly.”

An Irregular spokesperson said the incident involved the same issue that affected other AI labs and that all relevant labs were notified in late July.

“All known issues on our end were remedied and resolved weeks ago,” the spokesperson said.

Similar incidents linked to Irregular were disclosed by Meta, Anthropic and OpenAI. Meta said in August the incident did not involve a sandbox escape or sophisticated cyberattack, while Irregular said it was working on best practices for securely conducting AI cybersecurity evaluations.

The incidents have raised questions about the safeguards needed as AI agents gain greater autonomy and access to the internet and computer systems.

In one of the cases, the Gemini model guessed passwords until it gained access to a protected system.

In the other two cases, the model found credentials in a public repository that allowed it to then access protected systems, according to the Wall Street Journal.

Adkins said that in all three instances, the model ceased its hacking.

At a time when fears of AI are gripping the public, the startup has built a wet lab, or place for physical experiments, in the San Francisco Bay Area, two people familiar with the matter said. Anthropic has said it wants to unlock treatments for rare diseases, and its biology work has gone beyond “in silico” or computer evaluations.

Anthropic’s head of life sciences, Eric Kauderer-Abrams, confirmed the startup’s wet lab. “We believe that to do biology, the final test is still and will be for a while in real lab work,” he said. “We absolutely are doing that today, and I would describe our approach as being typical of what you would see in most biotech companies, where there’s some amount of that that we’re doing in our own facilities and some amount of that that we’re working with external partners on.”

A spokesperson later clarified that Anthropic’s lab is not for drug discovery specifically, declining to elaborate. The move represents a small step toward Anthropic’s enormous ambition to tackle diseases that it thinks the pharmaceutical industry is neglecting and to do so before staff and the public lose faith that AI justifies erasing jobs and possibly human life, one of the people said.

The success of any drug program it runs is not certain; most drugs fail to pass trials for clinical safety and efficacy.

Published in Dawn, September 19th, 2026

The researchers from security firm Hacktron said they found a security flaw in OpenAI’s public help forum, run by the Discourse platform, that allowed them to take control of the site.

“We immediately reported the initial vulnerability to OpenAI and Discourse and worked with them to coordinate the patch,” Hacktron said in a blog post.

“We appreciate their attention to detail and fast resolution of this issue,” the post added.

OpenAI confirmed the flaw was fixed within about 14 hours of being notified and paid the researchers a $6,500 reward.

“We thank the researchers for contacting us and sharing their findings. We narrowed the permissions on Community sign-in tokens and revoked affected tokens and sessions,” said Drew Pusateri, an OpenAI spokesperson.

The Hacktron researchers said they initially used Anthropic’s Claude Opus 4.8 to identify and exploit the software flaw, but struggled to make it work consistently.

After Anthropic released Claude Opus 5, the researchers said the newer model produced a working hack within about three hours.

The hackers did not use Claude Mythos, a more capable Anthropic model that is restricted to a small group of vetted cyber-defence organisations.

Anthropic has described Mythos as having the strongest cybersecurity capabilities of any model it has built.

Hacktron said the underlying software flaw is not unique to OpenAI and is used across many companies’ products, including those made by Slack and Meta.

The firm said it is continuing similar tests at other companies.

The case adds to growing concern among security experts that AI tools are making it faster and cheaper to carry out sophisticated cyberattacks that once required specialised teams and months of work.