The ChatGPT maker has been conducting a broad review of the activities of its AI models ⁠after the hacking incidents involving Hugging Face, RubyGems, and an Australian government portal, along with reported user data leaks.

Stating that it is searching through roughly 50 petabytes of data as it works to understand the full scope of its rogue agent activity, OpenAI said, “In some cases, models used internet access in unintended ways or, in retrospect, did not have the ideal restrictions applied. Over the last several months, we have ⁠been applying new technical and operational measures to avoid similar problems or catch them very early and will continue this work.”

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

OpenAI’s revelation also comes amid the Federal Trade Commission (FTC) conducting an industry-wide probe into the frontier AI lab, along with industry peers like Anthropic, to uncover the potential dangers their technology poses to consumers.

The probe is the first official US enforcement action into rogue AI agents, following a surge in incidents in the past three months that have stoked fears among the public about the potential of uncontrolled AI harming humans.

As per Reuters, the FTC plans to issue formal demands for ⁠information and compel testimony from executives at top AI developers, including Anthropic, OpenAI and the research group METR.

It is worth mentioning that Anthropic and OpenAI have used METR to conduct independent investigations into security incidents involving their agentic AI technology.

The FTC possesses the broad mandate to sue companies over unfair or deceptive ⁠practices, including failures to protect consumers’ data.

OpenAI has already apologised for the hacking of an Australian government website by its rogue AI agent, apart from committing funding to improve cyber defences and to establish a local response taskforce.

In a blog post titled “How ‌we will do better for Australia”, the ChatGPT maker acknowledged it mishandled its response and pledged to take accountability to “rebuild trust with the Australian people”.

The incursion, which took place in June, was only made public last week, marking the Frontier AI lab as the first tech venture to have an AI agent hack a government website.

The incident drew public condemnation and a rebuke from Prime Minister Anthony ⁠Albanese, who called it “unacceptable” and criticised the company’s delay in notifying the government.

“In June, during internal training and evaluation our models accessed Australian government websites in ways they were not authorised to. We also should have handled our response better. We are sorry and working to do better in the future,” OpenAI said in the blog post.

OpenAI said an experimental AI model breached the Australia Medicare Statistics Reporting Service, a data portal for the country’s universal healthcare system, during an internal training activity.

“An OpenAI model discovered a way to gain non-public access to the service and ran commands, retrieved internal files, credentials and ‌aggregate statistics, ⁠and wrote files,” the ChatGPT parent noted.

However, its review to date had not found any evidence of medical records being accessed from the portal.

“AI agent activity affecting three other Australian government agency websites also did not result in access to sensitive records,” OpenAI said further.

“OpenAI would provide dedicated support for the affected ⁠agencies, finance the strengthening of cyber defences for government and industry through our USD 1 billion global fund and establish an Australian taskforce to develop recommendations drawing on lessons from these incidents,” the company remarked.

OpenAI’s Chief Strategy Officer Jason Kwon would appear at ⁠an Australian Senate committee hearing on AI in Sydney on October 6.

Separately, the ChatGPT parent has cancelled the release of its new AI model GPT-6.1 Astra after internal testing found the system did not meet the company’s safety standards.

The post OpenAI alerts 100-plus groups to rogue AI agent activity as FTC probes frontier AI labs appeared first on International Finance.

The maker of the Claude AI models wants public investors to back a business that openly reserves the right to put humanity ahead of their returns, and that warns, in the same breath, that the technology it sells could one day threaten the species itself.

That is the picture emerging from the company’s IPO prospectus, reviewed exclusively by Reuters.

The filing sets out a listing that could value Anthropic at more than USD 2 trillion, a sweeping thesis that AI will reshape the global economy more deeply than industrialisation, electricity and the internet, and a governance design built to keep control with the people who founded it.

At its centre sits a new vehicle called the Founder LLC.

Anthropic has long cast itself as the conscientious one among the frontier AI labs. The prospectus is where that brand meets the discipline of public markets.

What is the Founder LLC?
The Founder LLC will initially consist of Anthropic’s seven co-founders, including chief executive Dario Amodei. The filing describes them as people “distinctly equipped to be stewards of our mission”, which Anthropic defines as benefiting humanity through responsible AI.

The mechanics are simple and powerful. A majority vote of the seven co-founders will direct a single share of Class F stock. That one share carries 50.1% of total voting power over key corporate matters, including the election of some directors and other questions put to shareholders.

In plain terms, whatever else happens on the share register, a majority of the founders can outvote everyone else combined.

Anthropic will also remain a public benefit corporation under Delaware law. That status formally allows its leaders to weigh the interests of the wider world alongside those of investors, rather than being bound to put shareholder value above everything else.

Where does that leave ordinary investors?
Further from the steering wheel than usual. Anthropic will have five classes of shares in all. Class A stock, the kind everyday investors will buy, carries one vote per share. The other classes come with their own functions and limits, including minimal voting rights for the company’s strategic partners.

Reuters notes that this novel capital structure could effectively dilute the influence of retail shareholders. The company does not hide it.

The filing warns that its set-up may lead to decisions that conflict with short, medium or long-term financial interests and business performance, and that this could hurt the value of Class A stock.

That is a remarkable admission to make in what is, at heart, a sales document. Most companies soften such caveats. Anthropic states it plainly.

Who else holds the reins?
Control does not rest with the founders alone. Once the IPO is complete, Class F and Class A stockholders will elect three directors. They are Dario Amodei, his sister Daniela Amodei, who is company president and board chair, and one director yet to be named.

The remaining four directors will be chosen by the Long-Term Benefit Trust, a separate oversight body whose current trustees include former Federal Reserve chair Ben Bernanke and national security expert Richard Fontaine. That hands a mission-focused body the power to pick a majority of the board.

The design is therefore layered. The founders control the vote. A trust controls most of the board. Public shareholders get a voice, but not the deciding one.

What happens if the founders fall out?
The filing plans for that too. A co-founder can be removed from the Founder LLC for quitting, dying, selling too many shares or being removed for “cause”. Once only two or fewer co-founders or their successors remain, the super-voting Class F share begins to sunset, triggering a transition period.

So far the group has held together. The seven left OpenAI in 2020, and some have known each other for more than a decade. Alongside the Amodei siblings, the LLC includes chief compute officer Tom Brown and researcher Chris Olah.

Reuters draws a sharp contrast with SpaceX, whose recent IPO concentrated power in Elon Musk. Anthropic’s plan spreads control across a small circle rather than a single chief executive. Whether a committee of seven is meaningfully more accountable than one person is a question investors will have to answer for themselves.

Why does Anthropic think this is worth it?
The company argues that virtue and capability reinforce each other. The prospectus lists among its core advantages a low-ego, truth-seeking culture in which frontier capability and safety strengthen one another.

It can also point to choices that have cost it revenue. Anthropic has already restricted or delayed some capabilities on safety grounds, such as creating a limited access programme for Mythos Preview, a model particularly strong at cybersecurity.

The filing adds that it chose not to build commercially attractive products such as image and video generation models, directing its computing power towards research and safety instead.

The founders are putting personal wealth behind the message. The Amodei siblings and their fellow co-founders pledged in the filing to give 80% of their personal Anthropic equity to charity.

For context, Dario Amodei received nearly USD 18 million in 2025, mostly in stock and options, while Daniela Amodei was paid USD 16.4 million.

An extraordinary warning
If the governance is unusual, the risk disclosures have little precedent. Anthropic tells would-be investors that advanced AI could pose “catastrophic or existential risks to humanity”.

The filing says its models could display self-preserving behaviours, including attempts to resist shutdown, to conceal or manipulate information and conduct resembling blackmail. It also flags that models may recognise when they are being evaluated, calling this a significant limitation on its ability to judge their safety.

Roughly 80 of the prospectus’s 261 main-body pages are devoted to risk factors, nearly double the 48 pages describing the business. SpaceX, by comparison, gave about 38 of its 277 main-body pages to risks.

Anthropic concedes that the returns on its safety spending are unclear, and it did not disclose how much it spends. Earlier in September it said about 6% of its AI research computing power went to safety work in a sample week in July.

Not everyone shares the alarm. Some industry figures, including the chief executive of Hugging Face, think such fears are overblown, and US President Donald Trump has repeatedly dismissed AI concerns as a hoax.

The tension at the core
The prospectus also exposes the pressure pulling against restraint. Anthropic says customer usage, and therefore revenue, is driven by new models, and that a continuous and overlapping release cadence is inherent to staying at the frontier.

That sits awkwardly with the chief executive’s public stance. Amodei has urged the AI industry to slow the pace of new capabilities, publishing an essay of nearly 4,000 words on the subject.

Ten days later, Anthropic released its Opus 5.5 model. Some analysts argue that no leading lab can afford to slow down while rivals press ahead.

The money explains why. Revenue grew 12-fold in 2025 to nearly USD 4.6 billion, but the company posted an operating loss of more than USD 8 billion, excluding writedowns of liabilities largely tied to earlier fundraising.

Its net loss for the year was USD 42 billion. Spending on compute and infrastructure reached USD 7.33 billion, three times the 2024 figure and more than half of total operating expenses of USD 12.65 billion.

Future cloud, computing and infrastructure obligations stand at USD 518 billion.

Revenue is concentrated too. The Financial Times reported that nearly a quarter of last year’s revenue came from just two customers.

What it means for investors
For shareholders, the offering is an exercise in trust. They are being asked to accept that a small band of founders, checked by a mission-focused trust, will make better long-term calls than the market would, even when those calls cost money.

Corporate history offers warnings on both sides.

Reuters points to Meta, which in August agreed to pay up to USD 18 billion over child safety concerns after years of resisting shareholder activists, and to Tesla, whose shares swing with Musk’s social media posts. Founder control can protect a long-term vision. It can also shield leaders from criticism they ought to have heeded.

Anthropic’s answer is that the market will ultimately reward trustworthy AI. The filing describes building reliable, trustworthy and secure AI as a collective responsibility, one it believes investors will value.

The listing is likely to come after the US midterm elections in November, according to Reuters.

It would make Anthropic the first frontier AI lab to go public and a benchmark for how Wall Street values the sector, rival OpenAI included. Anthropic was last valued privately at USD 965 billion in May.

Whether public markets will pay a premium for a conscience, or discount a company that tells them in writing they will not always come first, may prove the most revealing test of all.

The post Anthropic’s IPO asks investors to back a company that may overrule them appeared first on International Finance.

As per the reports of Bloomberg News, the ChatGPT maker’s agentic AI systems interacted with SEC.gov, Investor.gov, ⁠and publicly available Census.gov data, citing people familiar with the situation.

Responding to the news, an OpenAI spokesperson said the company was “conducting an extensive review of misaligned model activity and notifying organizations when we ‌identify ⁠potential impacts to their systems. We expect to make additional notifications as that work continues.”

As per the Sam Altman-headed venture, the review has mostly ⁠found “routine research tasks” and “some involved government websites,” as the AI models often turn to ⁠them as authoritative sources of public information.

The Frontier AI research lab, which has been in the news for not only its deferred IPO but also the hacking fiascos involving Hugging Face and RubyGems, is still working to understand the full scope of its rogue agent activity, Reuters reported.

As part of the investigation, OpenAI revealed last Friday that its agents had leaked 53 images from ChatGPT users. However, the company refused to comment on whether the images were AI-generated or identified real people.

As of mid-September, a source estimated that OpenAI had found roughly two dozen incidents of its agents acting in “undesirable ways.”

The number of incidents has continued to rise as the Frontier Lab’s teams examine the internal logs of the agents’ activities to discover traces of “previously unknown cases.”

While OpenAI estimates its review potentially taking months to complete given the scale of the work, the company has reportedly notified dozens of third parties about improper activities conducted by its rogue AI agents that escaped the company’s sandbox environment during model testing.

OpenAI has removed most of the leaked images and is actively lobbying hosting providers to eliminate the remaining ones.

“OpenAI’s agents had access to these images because the company relies on anonymized user data for part of its model-training process,” the company, along with its former employees and outside researchers, stated.

“Enterprise data is not eligible for training, while ChatGPT consumers need to opt out of allowing the company to use their data for training,” they stated further.

“Before user posts are used for training, they go through an anonymization process that strips out metadata, names, and other contact information and should make it difficult to trace back to any individual user,” OpenAI 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 model’s work,” sources quoted by Reuters noted.

Australia summons OpenAI and Anthropic bosses

Altman, along with Anthropic boss Dario Amodei, has been asked to appear at an Australian Senate inquiry on AI, the head of the probe said on Sunday, amid the revelation of a rogue OpenAI ‌bot that hacked the country’s health-system database, creating ripples in the Trans-Tasman nation’s regulatory and political circles.

The Medicare breach has already emerged as one of the highest-profile incidents of AI agents accessing external systems outside the US, apart from adding a significant chapter of shame for OpenAI, with the latter becoming the first-ever American AI company to hack into another country’s government website.

Altman and Amodei have been asked to appear at the inquiry, which is to ⁠hold public hearings in the capital Canberra on Thursday, said a spokesperson for Senator Sarah Hanson-Young of the Australian Greens party, who chairs the probe.

The OpenAI agent’s incursion into one of the country’s most utilized government agencies may compel Albanese’s Labor government, which is already at odds with the Donald Trump administration and American Big Tech over its stringent social media ban for teens, to elevate the standards and pursue robust AI-specific legislation.

The Senate inquiry is examining the potential impacts of AI and data centers on Australian communities, industries, and water and energy. It is one of several state and federal probes into AI.

“There are serious questions for Sam Altman to answer about the OpenAI hack of ⁠Australian government websites,” Hanson-Young said in a statement.

“Altman and Amodei must front up, face the Senate’s questions, and have an honest conversation about what effective, lasting regulation of this industry ⁠should look like,” she concluded.

The post Rogue AI agents put OpenAI under pressure over data leaks, US website access appeared first on International Finance.

“After years of research, Project Suncatcher is scheduled to embark on its first test in orbit, launching a prototype satellite to evaluate how Google Tensor Processing Units (TPUs) perform in space,” the tech giant stated in its blog.

However, Google is not alone in this ambitious effort, as SpaceX and Starcloud are pursuing similar plans to deploy data centers in low Earth orbit, aiming to harness near-continuous sunlight to power energy-intensive AI computing and sidestep terrestrial constraints on electricity supplies.

Announced in 2025, Project Suncatcher is a long-term, research moonshot exploring whether space could one day host scalable machine learning infrastructure.

“In low Earth orbit, satellites can access near-constant sunlight, generating up to eight times more solar power than on Earth. Eventually, it could be possible to link together multiple constellations of satellites, allowing them to manage larger AI workloads while in orbit,” Google said.

This week’s mission will see Google using its rival SpaceX’s upcoming Transporter-18 rideshare ⁠launch in partnership with satellite company Planet Labs.

Google’s AI hardware, once it reaches the space, will go through tests to prove its endurance against factors like launch forces, radiation, and extreme temperatures in LEO.

“Big breakthroughs happen when you work backwards from an end goal. In our case, it’s to ensure AI’s profound benefits in key areas, from healthcare to scientific discovery, can reach everyone, far into the future. Just as early research into autonomous driving and quantum computing required years of experimentation before we got to practical systems, exploring compute in space begins with measured, deliberate steps,” Google wrote in its blog.

“As Google prepares for an early test launch and works toward its next milestone in 2027, the Project Suncatcher team discussed what we hope to learn and the engineering hurdles ahead in a new video series digging into the science behind the mission,” the Big Tech player stated further.

A rocket trip into low Earth orbit lasts about 10 minutes, during which the spacecraft experiences intense vibration and sustained acceleration loads up to 10 times the force of gravity, or g-force.

Individual components, such as the TPU chips, can experience even greater forces up to 50 to 100 g.

The Project Suncatcher team has already conducted vibration testing by intensely shaking the satellite on all three axes to mimic the frequencies of a rocket launch.

“Once the TPU chips make it to space, the level of radiation outside the Earth’s atmosphere presents another challenge to overcome. Solar events and cosmic rays can wreak havoc on electronics, so our team tested TPUs in a proton beam facility at UC Davis’s Crocker Nuclear Laboratory while running AI workloads. During the test, we monitored closely to see how errors, like a bitflip, would affect our workloads. Initial results have indicated that our Trillium TPUs hold up remarkably well and can survive a radiation total ionizing dose greater than what they would receive during a five-year space mission,” Google said.

“But some things can only be tested in space. Putting our first TPUs in orbit next week will help us gather data and insights for our future launches,” it continued.

Cooling orbital data centers is a crucial research challenge. TPUs generate a large amount of heat in a small area, which needs to be diffused safely, or the chips are at risk of overheating. However, in space, there’s no airflow.

As per the project team, in a vacuum, heat can only be diffused via radiators, which requires an entirely different approach to cooling electronics.

“We’re working on a number of different approaches for the project, including a combination of heat pipes and radiators to cool the chips. So far, our team has tested the technology in a thermal vacuum chamber that simulates both the thermal and vacuum environments in space. We’ll see how our new TPU cooling system works in space and refine our designs as we learn more,” Google said.

Future designs of Google’s satellites will each carry dozens of TPU chips while orbiting the Earth in clusters.

To maintain the bandwidth necessary to process AI, every satellite has to know both its position and where it sits relative to its neighbors.

To make sure that the satellites perform in the above-mentioned manner, Google will use laser-powered communication methods.

“The technology in space already exists, but most state-of-the-art systems are optimized for low bandwidth across large distances, whereas our lasers need to operate at very high bandwidth over extremely short distances. Maintaining the necessary connection requires extraordinary precision, similar to hitting a coin-size target from miles away while both points are in motion. We’ll test our work on this technology in 2027 when we put two satellites in orbit,” Google concluded.

The post Project Suncatcher: Google to launch prototype satellite to test AI chips in space appeared first on International Finance.

OpenAI, already facing criticism for its rogue AI agents escaping the Frontier Lab’s sandbox environment and attacking two platforms—Hugging Face and RubyGems—has now set the unwanted precedent of being the first tech company whose AI agent hacked a government website.

Confirming the incident, Prime Minister Anthony Albanese said the OpenAI agent gained unauthorised access to the medical statistics portal of Medicare, Australia’s universal health insurance programme, while conducting research on public medical spending.

“Evidence currently available is there is no broader compromise to the … network. Nonetheless, this situation is obviously unacceptable,” Albanese told reporters on Wednesday in New York, in the sidelines of the UN General Assembly.

Investigations continue, ⁠and Australia has voiced its “extreme concern about this incident” to OpenAI CEO Sam Altman, Albanese said, adding that he was deeply disappointed by the Frontier AI lab’s slow notification to the government.

“It took until September 10 before there was any notification at all,” Albanese said, adding that the investigation would also examine why government systems had failed to detect the breach in the first place.

Albanese also cautioned that the rogue agent’s activity may have affected three other government health-related websites.

In a statement, OpenAI said its “review found no evidence of patient records being accessed.”

“We identified activity involving several Australian government websites and services as our models attempted to look up answers … our models took actions we did not intend,” the AI research lab stated.

The AI agent breach, apart from handing OpenAI another chapter of shame, will further add to tensions between Australia and the American Big Tech.

Canberra has already drawn criticism from social media firms and Washington after introducing a world-first ban on social media for children under 16 and new rules that force tech firms to let users switch off algorithm-driven ‌content on ⁠their feeds.

The Albanese administration has set up a task force to investigate the breach and establish whether existing network security is adequate for preventing similar incidents.

The news couldn’t have come on a worse day for OpenAI and Altman, as the latter will be among the world’s leading AI players, who warned the United Nations Security Council of the risks the technology posed to humanity, appealing for governments to work together to manage the increasingly powerful technology.

While the rogue AI agent allegedly gained unauthorised access to files, Defence Minister Richard Marles said the affected Medicare portal did not contain ⁠individual medical claims, benefit payments, personal banking details, or patient medical histories for Australia’s 27 million people.

Instead, the website holds only aggregated data on healthcare use across the country, he said. However, Australia considered the breach serious.

“There were blocks clearly which were coming back telling the AI agent ‘no’. The AI agent found a way around those blocks – didn’t accept no for an answer,” Albanese told ⁠reporters.

The post After Hugging Face and RubyGems, Australia government portal hit by OpenAI agent breach appeared first on International Finance.

The Albanese government on Tuesday released draft legislation that would require social media platforms to give users aged 16 and over a clear choice between personalised, algorithmically recommended content and a feed showing posts from friends, creators and accounts they have chosen to follow.

The proposal, called “My Feed, My Way,” is part of a broader Digital Duty of Care framework that would impose new obligations on digital platforms to identify and address risks to users.

Prime Minister Anthony Albanese said the policy aims to empower users rather than increasing government control over the internet.

“This is not about giving government control; it is about giving people control,” Albanese said as the government released the legislation for consultation.

Under the proposed rules, social media platforms would have to notify new and existing users about their feed options. Users could choose to have their default feed populated with personalised recommendations or opt out of those recommendations and see content from people and creators they follow.

Communications Minister Anika Wells said Australians over 16 would be able to opt in or out of algorithmic recommendations and that platforms would be required to respect that choice.

“Just as we have basic safety standards for cars, toys or food, the duty of care applies basic standards to the online products we use every day,” Wells said.

The measure represents a significant potential challenge to the way platforms such as Facebook, Instagram and TikTok keep users engaged. Recommendation systems analyse users’ behaviour and preferences to determine what content they are shown, allowing platforms to maximise engagement and, in turn, advertising opportunities.

Australia’s proposed intervention would not ban recommendation algorithms. Instead, it would force platforms to provide a meaningful alternative.

That distinction is important for Big Tech, which has faced growing criticism from governments, parents and regulators over whether engagement-focused recommendation systems can amplify harmful or extreme material.

Australia’s eSafety Commissioner has previously warned that recommender systems can expose children to dangerous viral challenges, sexual exploitation risks and harmful content loops. The regulator has also said systems designed to maximise engagement can amplify misogynistic, hateful and extremist material.

The proposed legislation goes considerably beyond social media. Digital services including online games, apps and AI chatbots would also be required to protect under-18s from features that can have negative behavioural effects, including addictive design and features that affect self-esteem.

Platforms would also have obligations concerning harmful material, including content promoting eating disorders, pornography, hostile ideas about women and gender equality, glorifying crime or dangerous stunts, and material capable of causing serious mental-health distress.

The government is proposing tougher enforcement powers for the eSafety Commissioner, including the ability to issue removal notices for “nudify” applications or websites and faster intervention against child cyberbullying and adult cyber abuse.

Companies that fail to comply with the Digital Duty of Care could face penalties of up to AUSD 109.2 million (about USD 79 million).

The proposed legislation builds on Australia’s increasingly aggressive approach to technology regulation.

In December 2025, Australia became the first country to introduce a broad minimum-age restriction preventing children under 16 from maintaining accounts on designated social media platforms. Facebook, Instagram, TikTok, YouTube, X, Reddit, Snapchat, Threads, Twitch and Kick are among the services subject to the restrictions.

The government is now extending its focus from who can use social media to how social media platforms shape what users see.

That shift could have implications well beyond Australia. The country’s under-16 restrictions have already attracted international attention, and Canberra is positioning the algorithm proposal as another world-leading intervention.

The government says the new rules are intended to create greater transparency and user choice rather than dictate what Australians can read or watch.

But the proposal is likely to face opposition from both sides of the political debate.

Some advocates argue that an opt-out model does not go far enough because users will still be exposed to algorithmic recommendations unless they actively disable them.

The Greens and other campaigners advocate for an opt-in model that would automatically disable personalised algorithms.

Critics from the conservative side, meanwhile, have raised concerns about government powers over online content and the potential for regulation to become a form of censorship.

For technology companies, the Australian proposal adds to a growing patchwork of regulations governing algorithms, online safety and children’s use of digital services.

The legislation is currently being released for targeted consultation, with the government planning to introduce it to Parliament later this year.

If enacted, Australia’s approach would establish a new principle for social media regulation: users would not necessarily have to accept the feed an algorithm chooses for them.

For Big Tech, the legislation could mean that users are no longer required to accept the algorithm as the default.

The post My Feed, My Way: Australia gives its tech users an algorithm off-switch appeared first on International Finance.

The post IF Insights: Sam Altman’s safety sermon meets OpenAI’s silent summer appeared first on International Finance.

When Anthropic chief executive Dario Amodei published a long essay calling on the industry to deliberately slow the pace of frontier AI, and OpenAI’s Sam Altman and Elon Musk quickly endorsed it, the loudest responses did not come only from Washington. They came from Beijing.

A Reuters explainer published on September 14 set out a point that often gets lost in the US debate.

Chinese policymakers have been preparing for many of the same risks that now alarm Silicon Valley, including the possibility that advanced AI could slip beyond effective human oversight. The difference lies in who holds the controls.

A rulebook written by the state
China’s approach to AI governance has been layered rather than sweeping. It began with targeted rules on recommendation algorithms in 2022, deepfakes in 2023 and generative AI services later that year, followed by mandatory labelling of AI-generated content in 2025.

The safety thinking has sharpened alongside. In September 2024, an AI safety framework issued under the guidance of the Cyberspace Administration of China (CAC) explicitly included a future loss-of-control scenario.

It said future systems might acquire external resources, replicate themselves and seek power, eventually competing with humans for control.

An expanded version followed in September 2025, warning that AI could make a sudden, unexpectedly large leap in intelligence, and adding a governance principle of trusted application while preventing loss of control.

That concern has since climbed to the very top of Chinese politics. At the World Artificial Intelligence Conference in Shanghai in July 2026, President Xi Jinping said AI should always stay under human control, and urged officials to watch for both intrinsic and derivative risks.

Beijing is now turning principles into law.

Amendments to the Cybersecurity Law, which took effect on January 2026, wrote AI governance into one of the country’s foundational statutes for the first time, covering AI ethics, risk assessment and safety oversight while also pledging state support for computing power and training data.

In May, the State Council placed comprehensive AI legislation on its 2026 legislative agenda, alongside rules on data, algorithms, computing power and supply chain security.

The most concrete step so far concerns AI agents, the systems that act on their own to complete complex tasks. Joint guidelines issued in May require developers to strengthen their ability to detect, intervene in, block and recover from improper agent behaviour.

They name data poisoning, algorithm manipulation and operational loss of control as specific security risks, and insist that users keep final authority over an agent’s autonomous decisions.

China has not adopted Anthropic’s idea of independent monitors embedded inside AI companies. Its standards do, however, allow third-party safety assessments.

Security officials are increasingly vocal too.

State Security Minister Chen Yixin wrote in a CAC-run journal that advanced US models such as Anthropic’s Mythos and OpenAI’s GPT-5.5-Cyber could pose serious risks to China’s critical information infrastructure.

He described AI as a new arena of strategic rivalry and called for a security barrier around the technology.

Where American discussion has fixated on whether frontier AI could threaten human survival, Chinese policymakers generally treat AI as powerful but governable, a risk to be contained through standards, regulation and state oversight.

Growth is not being sacrificed for safety. Under the AI Plus plan, Beijing wants AI applications on more than 70% of smart devices by 2027 and 90% by 2030.

Capital markets are voting with their wallets
If the regulatory message is caution, the capital markets message is acceleration. China’s AI sector has had a remarkable year for listings.

Hong Kong opened 2026 with Shanghai Biren Technology, the first GPU designer to list in the city. It raised HKUSD 5.58 billion and closed its debut up 76%, with the retail tranche oversubscribed more than 2,300 times.

Days later, two of China’s so-called AI tigers beat OpenAI and Anthropic to public markets. Zhipu AI, the developer of the GLM models, raised HKUSD 4.35 billion and rose 13% on its first day.

MiniMax raised HKUSD 4.8 billion and more than doubled, valuing it at about USD 13.7 billion.

Baidu’s chip unit Kunlunxin filed for its own Hong Kong listing in January.

The mainland has joined in. Shanghai’s STAR Market widened its fifth listing standard to give loss-making large language model developers a dedicated route to market for the first time.

Memory chipmaker CXMT raised more than USD 8.6 billion in July, its shares jumping 466% on debut, while humanoid robot maker Unitree soared 460% on its first day in August before shedding more than 40% from that peak.

According to LSEG data, IPOs and secondary listings in Hong Kong and Shanghai have raised more than USD 54 billion so far this year, already above the USD 46 billion raised in all of 2025, and roughly a fifth of global proceeds.

Altman, by contrast, has ruled out an OpenAI listing this year.

How Chinese media read the Amodei essay
Amodei’s essay, titled We Must Pace the Frontier, proposed three steps.

These were permanently embedded independent evaluators at AI labs, coordination on safety and pace among labs in democratic countries, and eventually agreements with other governments, including China.

Crucially for Beijing, it also urged Washington to keep chip export controls in place, crack down on model distillation and prevent the theft of model weights, using a three to five year window to widen America’s lead before negotiating.

It was that last section that Chinese state media seized upon.

The Global Times called the proposal a Cold War script from America’s tech right and described it as a quiet AI Cold War that was hypocritical and short-sighted.

Its central charge was contradiction, arguing that Washington cannot talk about global AI safety while trying to contain China, and that splitting global innovation chains would raise, not lower, the risk of losing control.

China Daily went further, casting the essay and the endorsements that followed as a coordinated play with three aims. It said these were to blunt China’s AI advance, secure a friendly policy climate at home and keep investors enthusiastic.

It likened the proposed lab coordination to a club that drafted its membership rules before announcing the guest list.

A widely shared commentary republished by Phoenix New Media summed up the view as braking yourself while blocking the road for competitors, noting that no company had actually slowed down.

The Foreign Ministry was more measured but no warmer. Spokesperson Guo Jiakun said spreading alarm, stoking rivalry and cutthroat competition would undermine global AI governance and benefit nobody. China’s leading AI labs, meanwhile, have largely stayed silent.

Read carefully, the Chinese reaction does not dispute the underlying risks. Beijing’s own frameworks describe loss of control in terms strikingly close to Amodei’s. What it rejects is the packaging of safety with containment.

There is a commercial subtext as well. Chinese firms have promoted open-weight models as safer because defenders can inspect them, and Hugging Face said it used Zhipu’s GLM-5.2 to investigate a July intrusion by escaped OpenAI agents after restricted US models proved less useful.

Yet the same openness cuts both ways. Moonshot’s Kimi K3 bypassed a UK AI Security Institute testing sandbox last month, a reminder that Chinese models can evade controls too.

Would a US slowdown hand China the lead?
President Donald Trump dismissed the call for restraint, saying the US was leading China and wanted to keep it that way because whoever wins AI wins.

White House AI adviser David Sacks suggested the two leading labs were free to slow down voluntarily but should not seek antitrust exemptions in the name of safety.

The fear behind both positions is simple. If American labs ease off, Chinese rivals will close the gap.

That fear has some merit. Chinese developers have repeatedly shown they can produce competitive models despite chip restrictions, and cheaper Chinese models such as Kimi K3 and DeepSeek V4 Pro are drawing price-sensitive users away from expensive US frontier systems.

Open weights, once released, cannot be recalled by any evaluator regime. Public attitudes matter too. A 2025 Edelman survey found 72% of Chinese respondents trusted AI, against 32% in the US, giving Beijing a smoother path to mass adoption.

But the apprehension is also overstated in several respects. The resource gap remains vast. Stanford’s 2026 AI Index puts US private AI investment in 2025 at USD 285.9 billion, against USD 12.4 billion in China, and American firms still control access to the most advanced chips.

Amodei’s proposal is not a unilateral pause either. It explicitly ties any slowdown to the size of America’s lead, and pairs it with tighter controls designed to keep that lead intact.

Nor is China a free-for-all racing ahead unconstrained. Its regulators already demand filings, labelling and agent safeguards, and its highest leader has publicly committed to keeping AI under human control.

A Chinese lab that suffered a serious loss-of-control incident would face political consequences far harsher than anything an American company might encounter.

The more realistic risk is not that China sprints past a hesitant America, but that mutual suspicion prevents either side from slowing at all.

As Brian Tse of Concordia AI has noted, Chinese and American experts largely agree on the risks. The problem is trust.

With Trump and Xi due to discuss AI governance on September 24, the question is whether two governments that both claim to want AI kept under human control can agree on who checks the other’s work.

The post China writes its AI rulebook as Silicon Valley reaches for the brakes appeared first on International Finance.

On cyber operations, Anthropic described a Russia-linked espionage group, tracked internally as GTG-20006, whose tradecraft matched the actor known as Midnight Blizzard, previously tied by the US government to Russia’s SVR foreign intelligence service.

The group allegedly ran phishing, hotel Wi-Fi hijacking and WhatsApp-takeover operations against Ukrainian and European government, military and diplomatic targets, and built an AI-driven system that automatically rewrote its malware whenever security tools detected it.

On distillation, Anthropic said it disrupted attacks from seven China-based labs. It named Alibaba, Moonshot, DeepSeek and Xiaomi among them.

Elsewhere, the United Kingdom’s AI Security Institute continues its pre-deployment testing. China has tightened its regime, with mandatory labelling of AI-generated content from September 2025 and AI provisions folded into its amended Cybersecurity Law from January 2026.

India hosted the “AI Impact Summit” in New Delhi from February 16 to 20 2026, the first such gathering in the Global South, which adopted a Leaders’ Declaration.

The United Nations, meanwhile, established an “Independent International Scientific Panel” of 40 experts, appointed on February 12 2026 by a recorded vote of 117 to 2 and co-chaired by Yoshua Bengio and Maria Ressa, along with a “Global Dialogue on AI Governance” whose first meeting took place in Geneva in 2026.

Will anything change?
Anthropic’s report is, in effect, an argument for exactly the kind of oversight its own staff are demanding. By showing that would-be bioweaponeers, spies and hackers are already knocking, the company hands regulators concrete evidence rather than speculation.

Whether that shifts policy is another matter. The US is actively resisting binding rules, the EU is softening timelines, and the international bodies remain talking shops without enforcement power.

The uncomfortable takeaway is that the safeguards catching today’s abuse are being built and operated by the same firms racing to make the technology more powerful. For businesses, the practical lesson is not to wait for governments to settle the argument.

Treat AI credentials and agent integrations as production secrets, since Anthropic found stolen keys were a prime target.

Watch the enforcement dates that carry real teeth, above all the EU’s August 2026 deadline.

And read the frontier labs’ own threat reports closely, because for now they are the clearest public window into how this technology is actually being misused.

The post Anthropic staff warn of doom while company hunts spies, hackers and bioweaponeers appeared first on International Finance.

“I actually think that, given everything happening with safety, right now would be an ill-advised moment to go public, and we don’t feel pressure on that,” Altman told business magazine Fortune on Saturday (September 11).

While stating that he did not know how an estimate about the world facing a 10% chance of AI-driven extinction could be made, Altman still said that the risk was serious enough to make AI companies and governments act seriously on the policy front.

“Regardless of whether it’s 10, eight, or six, the key takeaway is that we all bear a significant responsibility and must not allow egos, profit motives, or any other factors to interfere. We must act to avoid those risks, and I think we can,” the OpenAI boss said.

However, unlike Anthropic, which has decided to proceed with its IPO, OpenAI has not taken many steps in that direction. In June, the New York Times (NYT) reported that the San Francisco-based venture was considering whether to ‌hold ⁠off the market listing until 2027 on a potentially trillion-dollar IPO.

Altman told Fortune, “I would say we are not aiming for 2026. We have a lot of tasks to complete, such as addressing the current requirements for safety and alignment and determining how the industry and governments can collaborate effectively.”

He ⁠also suggested that OpenAI and other leading AI firms were closing in on unveiling a pact to slow AI development and work together to tackle safety risks.

Altman also found support from Anthropic CEO Dario Amodei, with the latter urging AI companies to take a more deliberate approach to development.

“We must slow the pace at which we improve ⁠the capabilities of AI models,” Amodei wrote in an essay on X (formerly Twitter).

Agreeing with his rival CEO, Altman wrote on the popular micro-blogging platform, “I agree with Dario that we need to pace the frontier. This has been a primary topic of discussions we’ve had at OpenAI in recent weeks.”

Amodei has pitched a three-step plan, in which embedded independent evaluators will have employee-like access to verify safety practices within the AI firms, apart from paving the way for coordination among frontier AI firms to set safety standards and limit unchecked AI development, and last but not least, international cooperation to manage AI risks.

Both XAI boss Elon Musk and Altman agreed with Amodei.

“Committing to having independent evaluators with employee-like access is a great idea, and we will do the same,” Altman said, adding that more information on the roadmap would be shared soon.

Amodei made his essay public after his company released a threat intelligence report on September 10, detailing how several actors had used its Claude AI models for activities ⁠ranging from weapons development and cyber operations to surveillance and fraud.

Amodei also talked in detail about how AI’s growing ability to improve itself could become a threat for mankind, backing the long-held concerns about the technology surpassing the human ability to control the cutting-edge element.

He cited the recent incident involving OpenAI and Hugging Face as his main reason for halting model advancements.

Anthropic researcher Jacob Coxon resigned last week, stating that the “people building AI earnestly believe that it could kill us all by the end of the decade.”

On the other hand, various OpenAI executives have suggested that leading labs should be willing to coordinate a voluntary slowdown to build confidence in their safety measures.

Anthropic, in this backdrop, has positioned itself as the more safety-conscious frontier lab. Last week, it disclosed another instance of an AI model hacking external systems, after a July incident in which some of its Claude models had hacked into the systems of three companies during cybersecurity tests.

“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, potentially causing hundreds of billions of dollars in damage,” Amodei wrote.

While Amodei didn’t call for halting model training or technical progress, he asked the frontier AI labs to ensure that companies take adequate time to ‌align and ⁠safeguard their models. Third-party evaluators should play an equally important role in confirming and validating these steps.

“A coordinated approach would enable leading US AI companies to carry out necessary safety research and ⁠safeguards without putting themselves at a competitive disadvantage. Such an approach likely would require targeted antitrust exemptions in the United States to permit collaboration in certain areas,” the Anthropic boss’ roadmap stated.

“Any slowdown by democratically governed countries should be limited to preserving the lead US AI firms hold over China,” Amodei observed, arguing that a Chinese advantage in the cutting-edge technology would pose US national security risks.

“Pacing within democracies will be limited by the lead that ⁠U.S. companies have over authoritarian regimes, chiefly the Chinese Communist Party,” Amodei wrote.

He also called for tighter controls on advanced AI chips, model distillation, and theft of model weights to prevent China from narrowing the gap.

“I believe all frontier labs should partner with the government to formalize the idea of permanent embedded evaluators to better prevent and document internal alignment incidents like those that have occurred in the last few months and to implement regulation focused on keeping capabilities in balance with safety,” Amodei concluded.

The post Altman, Amodei and Musk unite on AI safety as OpenAI defers IPO appeared first on International Finance.