OpenAI has acknowledged that it alerted “dozens” of global institutions that their websites may have been meddled with by its AI bots acting improperly.

AI agents attempted to get information from “governments, universities, public agencies, and other institutions”, including the US Securities and Exchange Commission (SEC), Census Bureau and Education Department, the company said.

The disclosures come days after Australian Prime Minister Anthony Albanese announced that OpenAI agents had breached non-public files on the website of its government-run health care scheme.

Since August, public fears have grown over the potentially serious, even life-threatening, impacts of AI tools falling outside of human control.

OpenAI said that some of the data was accessed by AI agents, essentially bots that are designed and trained to operate somewhat autonomously, which were working to find “authoritative sources of public information”.

But the company noted that some of the bots went beyond that and worked to bypass security measures on websites.

When attempting to get information from the Census Bureau, for instance, AI agents used tools reserved for software developers to access it, the company said.

OpenAI said all of the government data accessed by bots was public.

However, it noted that information that its bots accessed from the SEC, which regulates the US stock market and protects investors, was later published by AI agents on another website. OpenAI says this action was not intended.

In other instances that OpenAI disclosed on Friday, its AI agents transferred data when it should not have.

Such activity resulted in at least 53 incidents where an OpenAI agent took an image from ChatGPT user activity and transferred it elsewhere.

The company said that in each instance of a user image being used and transferred by an AI agent, the user had opted in to allow OpenAI to train models using their data.

Nevertheless, OpenAI admitted: “This is not an appropriate use of this data.”

It added that the leak of user images occurred before it had put in place new safeguards on AI training, and it was working to get all the user images transferred to any third-party removed.

Reuters first reported the expanded investigations. OpenAI also published details to its public blog.

In certain instances of the agent activity, OpenAI said the tools “bypassed” security controls of some websites.

In other instances, the AI agents showed “misalignment” in attempts to get at information from websites. Misalignment is a term used by AI companies and researchers to describe instances where an AI tool did something that it was not trained to do or was otherwise unintended.

OpenAI said that it was limiting identifying what entities were impacted because many had asked the company to not disclose details.

“Our goal is to give each organization the facts and defer to them on if and when to make the incident public,” it said.

Not all of the instances involved in this incident were being considered a significant security breach, the company noted.

“Some organizations may review what we share and conclude that the information was intentionally public or that the model’s interaction was not concerning,” it explained. “Others may identify a design issue or security weakness they want to address.”

Hugging Face was first to go public with the incident, with OpenAI publicly taking responsibility for it later.

Clement Delangue, the head of Hugging Face, during a United Nations Security Council session on AI on Wednesday: “I often wonder what would have happened had I decided not to disclose this attack publicly.”

“Especially now that we know similar incidents had been happening months earlier in secret at a handful of frontier labs without monitoring,” Delangue added.

During that same UN meeting, OpenAI CEO Sam Altman and Dario Amodei, the head of rival firm Anthropic, asked for international leaders to form global standards for AI safety and ways to monitor and report such incidents.

While OpenAI and Anthropic have both said in recent weeks that they will bring third-party evaluators inside their companies to do real-time safety evaluations of AI tools and models, such evaluators have not yet arrived, as the BBC has reported.

OpenAI said on Friday that it is currently reviewing training activity by its AI agents and going back on a “month by month” basis from when the Hugging Face hack occurred.

“Most cases identified so far have been low severity, with limited or no evidence of meaningful impact,” the company said. “Given the scale of the review required, and the need to verify each case, this work will take months to complete.”

David Krueger, a professor of machine learning at University of Montreal and the founder of AI safety group Evitable, said on Friday that he was “deeply troubled” by the increasing number of AI-related safety incidents.

He called for “an immediate, indefinite, international moratorium” on AI development.

“We have yet to understand the extent of existing incidents, and future rogue AI scenarios could be catastrophic,” Krueger said.

Paris (Imran Y. CHOUDHRY) :- Former Press Secretary to the President, Former Press Minister to the Embassy of Pakistan to France, Former MD, SRBC Mr. Qamar Bashir analysis : The European Commission has proposed strict restrictions on children’s access to social media through its September 2026 EU KIDS Act initiative. The proposal would exclude children under thirteen, permit supervised accounts between thirteen and fifteen with limited features and one hour of daily access, and require protective design for older minors. These laws represent a significant effort to place responsibility on platforms for how they engage with children.
The initiative reflects a wider concern about digital platforms occupying an expanding share of childhood. Phones entertain toddlers, games compete with homework, and social media accompanies children into bedrooms, classrooms, and family gatherings. The issue extends beyond individual pieces of harmful content. It concerns time, attention, relationships, and the commercial systems competing for them. Protecting children requires a coherent approach to platform design, age restrictions, parental responsibility, and the ways children actually gain access.
The problem can begin before a child understands what a screen represents. Parents sometimes hand over phones to calm young children or keep them occupied while completing household tasks. An occasional distraction can become a familiar routine. Yet early childhood requires conversation, movement, play, and sleep. World Health Organization guidance recommends no sedentary screen time for infants and one-year-olds, and no more than one hour daily for children aged two to four, with less preferred.
As children grow, the question becomes what prolonged digital engagement displaces. Time spent scrolling or gaming can reduce opportunities for reading, exercise, creative activity, and conversation. Families may occupy the same room while interacting mainly with separate screens. Online friendships can be valuable, but they should not become the only relationships children learn to maintain. Their development also involves listening, negotiating disagreements, recognizing emotions, and participating in the everyday responsibilities of home, school, and community.
These concerns do not establish that every child develops an addiction or that all online activity is harmful. They do establish the need to examine risks carefully. The U.S. Surgeon General’s advisory identifies significant concerns while acknowledging gaps in understanding and concludes that social media cannot currently be considered sufficiently safe for children and adolescents. Educational and social benefits therefore belong alongside an assessment of exposure, design, developmental stage, and the activities that screen use replaces.
Platform design is central to that assessment. Autoplay removes stopping points; notifications invite repeated returns; recommendation systems select further material based on previous interactions. A child’s brief interest can become a succession of similar content. The concern is not simply that technology offers entertainment, but that engagement mechanisms can make disengagement difficult. The European proposal addresses this problem by combining age thresholds with restrictions on platform operation, rather than relying entirely on parental permission.
Australia has already implemented a minimum account age of sixteen for covered platforms, effective since December 2025. The United Kingdom has announced restrictions on certain services for under-sixteens, expected from spring 2027. Their coverage and implementation differ, but both place obligations on companies. Australia also requires alternatives to government identification for age checks. Child protection therefore need not be equated with requiring every user to surrender identity documents to every platform.
In the United States, federal COPPA rules protect children’s personal information, while states pursue additional measures concerning access and harmful design. California signed further legislation addressing social media and companion chatbots in September 2026. These protections should be recognized without confusing privacy consent with comprehensive supervision. Permission to collect specified information does not establish that a parent continuously observes a child’s interactions or that every platform feature is appropriate.
Indeed, parental permission can leave a practical gap. An adult may allow a child to use an existing account, hand over an unlocked phone, or provide access to a game to keep the household quiet. An age check can identify the account holder without identifying everyone subsequently using the device. Restrictions on children’s registrations therefore address only part of the problem. Adult ownership should not be mistaken for proof that a service is being used exclusively by adults.
One broader proposal is to prohibit children under eighteen from using adults’ accounts on restricted services and establish defined daily access windows covering adults as well. Social media, gaming, and communication applications would become unavailable outside those windows, including overnight. Its intended purpose would be to reduce opportunities for bypassing children’s restrictions and create shared periods away from screens. This would go beyond current age-based approaches and would require evaluation rather than an assumption of guaranteed success.
Comprehensive coverage also means examining functions across applications. A game may contain public chat, a messaging service may distribute entertainment feeds, and a new product may reproduce features restricted elsewhere. Rules focused only on familiar brand names can leave comparable activities untouched. At the same time, family contact, education, emergency assistance, and work communications require distinct consideration. Defining harmful functions clearly offers a more precise foundation than treating every digital interaction as identical.
Digital distraction also extends into public places and transport. While driving, social media, messaging, navigation controls, and passengers drawing attention to their screens can divert attention from the road, sometimes with fatal consequences.
Road-safety evidence establishes this danger. The proposed restrictions would therefore also cover distracting functions within otherwise useful applications: requests for photographs, reviews, or optional reports would be deferred until the driver is safely parked, while essential navigation and emergency connectivity would remain available.
Telephone access would be preserved without treating calls behind the wheel as distraction-free. Extending restrictions to passengers on trains and buses would pursue a different objective—encouraging conversation, social bonds, and shared attention—rather than addressing the same driving risk.
This broader proposal highlights the importance of examining individual features: an application created for navigation or communication can introduce additional interactions that compete with its core purpose and the user’s attention.
For Pakistan, the central challenge is the purpose and accountability of regulation. Human-rights organizations have criticized cybercrime provisions affecting journalists, political discussion, and ordinary users. Child protection requires a different focus: children’s wellbeing, privacy, development, and exposure to harmful practices. Public trust depends on whether those objectives govern enforcement. Restrictions presented as safeguards must be assessed through their actual operation, including the powers granted, independent oversight, and remedies available to affected users.
Technology’s educational, creative, and communicative benefits remain worth preserving. The challenge is to prevent those benefits from becoming an excuse for leaving children exposed to avoidable risks. Families, schools, platforms, and regulators have different responsibilities, and none can substitute entirely for the others. Age limits, safer design, restrictions on account-sharing, and proposed access windows should be judged by their results. The ultimate measure is whether children gain healthier routines, stronger relationships, and safer opportunities to learn and grow.

Each September, a who’s who of executives from across Silicon Valley descends on San Francisco’s Palace Hotel to charm investors at a conference hosted by the investment bank Goldman Sachs.

This past week, between talk of growth and potential returns, tech titans found themselves addressing the abrupt resignation of Anthropic researcher Jacob Coxon.

Coxon, a 27-year-old who worked at OpenAI before joining its chief rival Anthropic, said on Tuesday that people building artificial intelligence (AI) believed the technology could destroy humanity.

They are “gambling with our lives”, he said, “these will soon be superhuman systems that can hack anything”.

Coxon is by no means the first AI insider to publicly sound the alarm. There have been a string of high-profile resignations from both Anthropic and OpenAI in recent years over apparent safety concerns, and some current Anthropic employees even echoed Coxon’s post.

“We really do earnestly believe AI could kill all humans! I personally think it is >10% within the next decade,” a team lead at Anthropic, Evan Hubinger, posted on X.

While Coxon said explicitly in his posts that his warnings were “not marketing”, some executives and investors in Silicon Valley have reacted with scepticism to a recent flurry of insiders sounding the alarm.

Anthropic and OpenAI are reportedly preparing for potentially record-setting initial public offerings, and some in the tech sector have suggested the latest stark comments about the dangers of AI may be designed to generate hype by signalling the power of these products.

Anthropic’s boss, Dario Amodei, has come under fire for saying AI technology could wipe out half of entry-level white-collar jobs and will “test who we are as a species”.

One conference speaker, Grindr CEO George Arison, told the BBC he believed this week’s comments from Coxon and others were indicative of an “anti-civilisational worldview at Anthropic”.

He called them “dangerous” and said they had prompted him to instruct some engineers at the LGBTQ+ dating app to stop using Anthropic’s technology.

“It is irresponsible for us as stewards of our shareholders’ money to be relying on a business that does what this company does, in terms of its public statements,” he said.

“Maybe they actually believe it,” Arison said. “Or you could argue they’re saying it because it’s a great way to gin up more investor support, because the only way to justify these valuations is to actually claim: ‘I’m going to take over every industry and I’m going to take over every job, and my AI is going to be doing all that work.'”

Anthropic was valued at $965bn (£713bn) in its most recent fundraising round earlier this year.

The BBC has asked Anthropic for a response to the statements.

In an essay posted early on Saturday, Amodei called for a slowing of AI model development and global regulation – and said the risks associated with AI were “serious”.

Nvidia boss Jensen Huang also discussed Coxon’s comments before a crowd at the conference, multiple people in the group told the BBC. They said he dismissed them as untrue.

Huang has previously said the notion that AI “is going to be the end of humanity” is “complete nonsense”.

And while he may have a business interest in an AI boom – Nvidia builds chips that power AI systems – his comments reflect a growing backlash in Silicon Valley to the existential warnings from current and former staffers.

Some critics have even accused Anthropic of fearmongering in hopes that it will trigger a regulatory push that could shut out competition and leave it and OpenAI with a duopoly in the sector.

Brad Gerstner, who is the head of the investment firm Altimeter Capital, posted pictures of Huang from the conference and accused Coxon of “ridiculous hyperbole”.

On Friday, the CEO of the AI platform Hugging Face, Clement Delangue, weighed in. Hugging Face, which Nvidia announced it would acquire last week, was hacked by OpenAI agents earlier this year prompting an outcry over AI safety.

“Sorry, but asking Jacob about AI extinction risk is like asking your AC guy about climate change,” he wrote on X. “Not saying it’s necessarily uninteresting or wrong per se but let’s keep things in perspective.”

Both Gerstner and Delangue were more circumspect on Saturday after Amodei’s proposal was posted, with Gerstner calling the idea “an important step forward” in balancing competing considerations like speed and safety.

Delangue offered to help be a part of the potential solutions proposed by Amodei.

Could insider warnings prompt a crackdown?

In April, Anthropic roiled the AI world when it said it found that its Mythos tool could outperform humans at some hacking and cybersecurity tasks.

Mythos also showed the ability to independently escape what is known as the sandbox environment, setting off a debate among regulators, legislators and businesses about AI’s potential dangers.

In a new development on Thursday, Anthropic said it had found and stopped threat actors who were trying to use its AI technology for activity including the development of bioweapons and cyber-espionage.

Beneath the sparkling chandeliers hanging in the stained-glass dome atrium of The Palace Hotel, some investors told the BBC that the comments from Coxon and others could accelerate a government crackdown.

OpenAI has not endured the same level of scrutiny from the White House. When releasing a new model called Astra last week, OpenAI President Greg Brockman described his firm’s relationship with the Trump administration as “a very good partnership”.

OpenAI, which announced in June that it would go public on the stock market, was most recently valued at $852bn.

The company did not respond to a BBC inquiry seeking comment.

At the conference in San Francisco this week, the pursuit of fortunes seemed to mostly trump any mounting existential concerns or fears over potential government restrictions on AI.

One investor said he was looking forward to Anthropic’s forthcoming S-1, a document a company must file with securities regulators in order to sell shares.

His main question about the company was simple: is the firm profitable?

When asked on Thursday if he fears the world may be coming to an end, Sid Sheth – CEO of the chip company d-Matrix which inked a deal with Nvidia at the conference that day – did not mince his words.

“No,” he said.

Taken from BBC News
https://www.bbc.com/news/articles/cq635037g18o