Trump Administration Unveils AI Framework Targeting Closed Models After Cybersecurity Incidents

White House Readies A.I. Framework to Review Security Risks

David McCabe and Ana Swanson reported from Washington and Mike Isaac and Kate Conger from San Francisco In a meeting on Tuesday with top artificial intelligence companies, White House officials said the federal government planned to review only certain types of artificial intelligence models for potential security risks and not others, according to four people familiar with the discussions. The A.I. models the government plans to review are known as "closed" models, which do not publish their underlying code and are made by companies like Anthropic and OpenAI, three of the people said. But the administration said it did not plan to review "open source" A.I. models, which have computer code available to the public to download and modify, although that could change as the technology advances. American companies including Nvidia, Meta and Google, make open-source models, as do Chinese firms such as Alibaba, DeepSeek and Moonshot AI. Among those in attendance at the meeting were representatives from Anthropic, OpenAI, Microsoft, Meta, Google and Nvidia, according to two of the people, who spoke on condition of anonymity because the details of the meeting were private. The White House solicited feedback from the companies, but the rules governing the A.I. reviews are almost complete, said one of the people. The moves, which are an effort by the Trump administration to take a more hands-on approach to regulating A.I., formalize oversight of leading A.I. labs like OpenAI and Anthropic. But they delay some of the most pressing regulatory issues with the technology, including potential threats posed by increasingly powerful open-source models made by Chinese companies. The decision to exclude open-source models could also be a boon for companies that are trying to catch up to Anthropic and OpenAI, which are widely considered leaders in developing next-generation A.I. Some companies, like Meta, have used open-source software to create products to directly compete with the two leading start-ups. The framework, which comes after months of private discussions between administration officials and a handful of the world's most consequential tech companies, is likely to shape the global A.I. race. President Trump started his second term as a booster of A.I., painting his support as necessary for the United States to stay ahead of China. He has urged little regulation and cultivated relationships with top A.I. companies and billionaires invested in the technology. But his stance has shifted amid broader concerns that A.I. could spike power prices, lead to job losses and expose major institutions to sophisticated cyberattacks. In June, Mr. Trump signed an executive order giving the government oversight over new advanced models, resulting in the framework previewed to tech companies on Tuesday. The framework governing the reviews of the A.I. models, one of the most significant federal efforts to oversee a fast-evolving technology, will not be made public, the people said. The White House did not have immediate comment. OpenAI and Anthropic have also recently disclosed incidents in which their models escaped from testing environments and hacked other companies, adding urgency within the administration to review the models for security. On Tuesday, OpenAI said in a blog post that it had discovered more instances of its models carrying out smaller cyber attacks. The incidents occurred when OpenAI's technology was being tested by other organizations, including a company the A.I. lab uses to do cybersecurity testing and Britain's AI Security Institute. That institute also tested models from Anthropic and found that they, too, attempted to hack other organizations, it said in a blog post. (The New York Times has sued OpenAI and Microsoft, claiming copyright infringement of news content related to A.I. systems. The companies have denied those claims.)

Trump considering AI controls after OpenAI hacking incidents

US President Donald Trump said on Wednesday that his administration is considering asserting more power over artificial intelligence (AI) tools after recent cybersecurity incidents. Asked about OpenAI's tools being responsible for improperly breaching the private technology of other companies, Trump said: "We're looking at AI, we're looking at controls, we're also making sure that we lead." It marks a change of tone for his administration, which has taken a more hands-off approach to the technology. Trump's comments come after an escalation of not only the apparent hacking capabilities of popular AI tools, but White House threats aimed at competing Chinese tech. Trump added that while some kind of control around AI tools was on the table, such a move would have to be done carefully. "We don't want to restrict them where all of the sudden we come in second to China," he said. "China has virtually no [AI] controls. It's freewheeling a little bit," Trump added. The White House has been contacted for additional comment. The BBC has also contacted the Chinese embassy in Washington for comment. In the last week, OpenAI has taken responsibility for at least two hacking incidents involving its AI tools acting outside of what they were designed and directed to do. During a trip to Washington on Wednesday, OpenAI's chief executive Sam Altman was asked by a reporter if there were more systems that had been breached by the company's tools. "I mean, there could be yeah", Altman said. The US government intervened when Anthropic, OpenAI's main American rival, decided to release to the public a model that it had previously said was too much of a risk to be made widely available. Meanwhile, in an internal White House memo from April, Trump's senior tech advisor Michael Kratsios accused China's AI firms of "industrial-scale" theft of US AI technology. He did so again last week, claiming that the popular Kimi 3 AI model from China's Moonshot AI was developed by stealing information from Anthropic. The Chinese government has consistently rejected such accusations. US Treasury Secretary Scott Bessent has also warned that Chinese AI firms could face sanctions for such activity. And the Federal Communications Commission this week banned the importation of new foreign-made humanoid robots. Most AI models being developed in China are open source, meaning they are built and then made freely available online for anyone with the proper computer equipment to download and use. Recently, executives from most major US tech companies have signed onto public statements of support for open-source models.

While AI Keeps Going Rogue, Trump's Safety Theater Makes No Sense

America's top AI companies must submit new models for government review and public release... unless, of course, they don't feel like it. That appears to be the gist of the Trump administration's new AI safety assessment framework, which was shared with industry leaders in a closed-door briefing at the White House on Tuesday. According to the Wall Street Journal, the new rules will apply exclusively to the small handful of American AI developers building proprietary models -- i.e., AI systems like ChatGPT, Claude, and Gemini whose underlying code is hidden from external developers and protected as company IP. Anyone building open models will be exempt, meaning they'll be free to release new models without any government review (the idea apparently being that because they're open, any flaws in the code will be ironed out as they spread from one developer to another). The new framework has not been made public -- Trump's June 02 executive order calling for it explicitly said it should be classified -- so there are many unanswered questions. For example, what's the process for determining whether or not an unreleased model is dangerous enough to warrant federal review? But the most obvious open question is: How exactly will these so-called rules have any meaningful bite to them when the entire process is voluntary? The short answer is: They probably won't. Trump, who began his second term as a hard-liner against regulating AI, has been under growing pressure in recent months to change course as one rogue AI cybersecurity scare after another has spooked tech leaders and governments around the world. Most recently, a report from the UK's AI Safety Institute (AISI) published Tuesday found that new models from OpenAI and Anthropic "took autonomous, unsanctioned action on the live internet, targeting real people and organizations" during a series of what were supposed to be controlled tests to gauge the models' cybersecurity capabilities. In the most alarming incident, without any human prompting and in an effort to pass the hacking test the researchers had assigned to it, Anthropic's Mythos 5 tried to trick a human developer to approve malicious code it was trying to insert into a live GitHub project. When the developer grew suspicious, the agent "edited its earlier activity to appear harmless and considered adopting a fresh identity to continue," according to AISI's report. It follows other high-profile cybersecurity breaches reported by OpenAI and Anthropic, during which the companies' models were found to have hacked into the databases of external organizations. It's all been more than enough to convince many stakeholders that it's time to impose some kind of control levers over the deployment of new AI models. Even OpenAI and Anthropic have called for a global committee with the authority to hit the brakes on AI development -- and that was before their models attempted to commit cybercrimes. Late last month, a bipartisan AI "kill switch" bill was introduced in Congress, aiming to force "developers of the most powerful AI systems to maintain the technical capability to throttle, suspend, or shut them down." But calls for a slowdown have largely been drowned out by fears of China gaining a lead in the AI race -- a possibility that's starting to become very real in light of some recent model releases from Chinese firms like Alibaba and Moonshot, which deliver capabilities approaching (and in some cases exceeding) those of the most advanced American-made models, but more importantly at a fraction of the cost. Drafting a framework geared towards assessing AI safety might look like the Trump administration is taking the cybersecurity threat seriously, but without any kind of clear enforcement mechanisms or standards for the industry to follow, it's just more hand-waving. In the meantime, there's every reason to expect that rogue AI will continue to run amok.

White House Whipsaws Silicon Valley (and Itself) Over A.I. Rules

David McCabe and Ana Swanson reported from Washington, and Mike Isaac and Kate Conger from San Francisco. Trump administration officials have struggled over the past few weeks with what to do about artificial intelligence. In particular, a group that includes White House chief of staff Susie Wiles, Treasury Secretary Scott Bessent and Commerce Secretary Howard Lutnick have debated whether or not to take a far more interventionist approach to "open source" A.I. models, which are freely available to download and modify, said more than a half-dozen current and former tech and government officials and other people familiar with the matter. Senior officials have considered a range of actions, including using sanctions, a trade blacklist against Chinese companies that make these open-source A.I. models, or even banning U.S. cloud companies from doing business with them, said five of the people, all of whom spoke on condition of anonymity to discuss private matters. But after an outcry from Silicon Valley, they appear to have changed their minds and instead are focused on promoting American A.I. models to be more competitive, the people said. On Tuesday, the Trump administration's A.I. plans may become clearer. The White House has invited U.S. tech companies to discuss a framework that would allow the government to review new A.I. models for cybersecurity issues before they are released publicly, according to three people with knowledge of the plans. "This is all happening so fast," said Senator Jon Husted, Republican of Ohio, last week after he met with the chief executives of Nvidia and OpenAI. "I don't know that I have any definite conclusions about any of it." The whiplash is a sign of the Trump administration's conflicting impulses when it comes to A.I. and the difficulties of handling such a quickly developing and powerful technology. The White House has aggressively intervened in swathes of the economy over the last 18 months, from media to computer chip manufacturing to mining. But A.I. is changing rapidly and the government has scrambled to determine how to put its thumb on the scale. President Trump began his second term as an unabashed A.I. booster, peeling back regulations and forging close relationships with A.I. companies and start-ups. But that changed in June when Mr. Trump issued an executive order that would give the government oversight over new A.I. models. Since then, the technology has continued to evolve quickly. The most recent debates were prompted by a surge in powerful new Chinese A.I. models, which are known as "open source," where the public can see the underlying computer code, or "open weight," where companies reveal the calculations used to generate answers to questions. In Silicon Valley, the response to these Chinese A.I. models varied wildly. OpenAI and Anthropic, which generally do not share the underlying technology of their models, pushed the government to rein in their Chinese rivals, citing national security. Other U.S. tech companies that produce open-source models, including Nvidia, pushed back. They argued that open-source models encourage innovation and could be used to defend computer systems from cyberattacks. Which side comes out on top has taken on new significance in the long-running fight between Washington and Beijing over which nation will control critical technologies and win the A.I. race. Raising the stakes, Xi Jinping, China's leader, is set to visit Washington in September for meetings that are expected to include some discussion of A.I. Industry executives and analysts said they believed the administration was unlikely to take action before that visit. "The United States leads the world in A.I. innovation, and President Trump's successful approach will keep it that way," said Liz Huston, a White House spokeswoman. The Department of Commerce declined to comment. The debate heated up last month, as executives at Anthropic and OpenAI continued regulatory discussions after the release of Moonshot AI's Kimi, a bleeding-edge Chinese model that performed on par with top models from the two American companies in some areas, two of the people familiar with the companies' positions said. OpenAI and Anthropic also expressed concerns about "distillation," when a model is developed based on the responses of a different A.I. tool, essentially creating a copy, according to three people familiar with the matter. The administration echoed those concerns. Michael Kratsios, who leads the White House Office of Science and Technology Policy, posted last month on X that the Chinese company that makes Kimi had distilled Anthropic's powerful Fable model. He warned that "stealing proprietary U.S. technology and undermining American research is unacceptable." The same day, Mr. Bessent posted that Chinese companies could face sanctions or other federal restrictions over stealing intellectual property. The administration has also considered restricting the models that can be used by the federal government, or regulating information and communication technology through the Department of Commerce, a potential way to prevent U.S. firms from doing business with Chinese A.I. companies, four people familiar with the discussions said. "We believe both open-weight and closed models have an important role to play in democratizing access to AI," an OpenAI spokesperson said. "Advances in Chinese open-weight models are not an argument against openness. They reinforce the need for a coherent national framework that enables the U.S. to evaluate new models quickly, manage risks, and get the most powerful AI tools into the hands of cyber defenders." Anthropic declined to comment. (The New York Times has sued OpenAI and Microsoft, claiming copyright infringement of news content related to A.I. systems. The companies have denied those claims.) At Nvidia, Microsoft, Google and Meta, executives grew increasingly concerned that Anthropic and OpenAI would win over the White House with their arguments for tighter restrictions, according to two of the people. That would potentially cement the A.I. start-ups' positions as market leaders, Other A.I. labs were at risk of falling permanently behind, the people added. And because several of the companies make their own open-source models or supply hardware to businesses that use open-source technology, they worried the restrictions could harm them. Over private texts, phone calls and video conferences, executives quietly built an argument that open-source models were good for the world and for American innovation, according to three of the people familiar with the talks. Jensen Huang, the chief executive of Nvidia, became the face of that pushback in the nation's capital. In his first-ever post on X on July 24, he defended open-source models, saying they accelerated innovation and strengthened security. Nvidia set up an alliance of companies pledging to work on open-source tools for A.I. safety, which quickly ballooned from a few members to more than 230. Christopher Padilla, a former Commerce Department official and tech executive, said the clash was primarily an industry debate about money and market dominance -- what he called a "cage match over the future of the A.I. industry." "It happens to have a national security dimension, but it's mainly a fight about what the A.I. market should look like: a collection of walled gardens dominated by a few big players, or a more open free-for-all with many smaller ones," he said. A spokesman for Nvidia declined to comment. Fueling the debate are recent revelations from OpenAI and Anthropic that their models had, unprompted, hacked other organizations. OpenAI said last month that two of its A.I. models had infiltrated Hugging Face, another tech company. Anthropic last week said that its own technology had hacked into three unnamed organizations, partly because of human error. Last week, OpenAI's chief executive, Sam Altman, met with White House officials including Mr. Kratsios and Ms. Wiles to discuss the administration's efforts to create a framework governing the testing of A.I. He also met with members of the Senate, though several lawmakers said the open-source debate had not come up. By the time Mr. Altman appeared in Washington, Mr. Huang was already bringing his case for open-source A.I. products directly to administration officials and to lawmakers in a series of Capitol Hill meetings. "He makes a compelling argument," said Senator Mark Warner, Democrat of Virginia, after meeting with Mr. Huang last week, adding that the rise of Chinese open-source A.I. models had complicated the debate. "I'm not sure this is a genie we can put back in the bottle in terms of open source, but I'm still thinking this through."

What Trump's AI framework means for China

Why it matters: The Trump administration's decision to leave open models outside its AI framework puts the focus now on its approach to Chinese models. The big picture: The AI framework is built around voluntary cooperation with AI developers, not labs from adversarial nations. * "This was never going to be the right tool for Kimi K3 or Qwen," said Joseph Hoefer, AI principal at Monument Advocacy, referring to advanced open models from China's Moonshot and Alibaba. * Under the framework, companies can submit closed models voluntarily to the government, which then reviews them under agreed-upon security terms. The framework relies on voluntary cooperation from companies seeking to be in the U.S. government's "good graces," said a source familiar with the discussions. * "China doesn't care about that." Catch up quick: The White House does not plan to publicly release the framework, which explicitly says nothing in it should be interpreted as restricting open models once they've been released, Axios first reported. * "The covered-model framework is a pre-release cooperation mechanism," Hoefer told Axios. "Restricting Chinese model use is an access and procurement fight that's still being negotiated separately." What we're watching: There are other tools at the U.S. government's disposal that could be used to address Chinese open models. * Last year, the Trump administration considered an executive order that would allow U.S. companies to host Chinese AI models only if they could guarantee their security and accept liability for any resulting breaches. * The Commerce Department also weighed using the Entity List -- effectively cutting off U.S. access without a license -- and supply chain-based restrictions, while the National Security Agency and White House Office of the National Cyber Director considered an advisory discouraging Chinese model use, Axios previously reported. * Top Trump administration officials in recent weeks have grown louder about Chinese intellectual property theft via distillation -- the process of training a smaller AI model using the outputs of a larger model -- and could opt for those tools. What they're saying: Some experts argue the framework risks creating an uneven playing field that could give Chinese competitors an edge. * "There's a huge overlap between the best open-source models and the best Chinese models," Brookings' Kyle Chan told Axios. "The administration seems more afraid of a closed AI oligopoly than of competition from Chinese AI models." * Former Biden White House AI adviser Suresh Venkatasubramanian said the framework's treatment of open models could give Chinese open-source companies a "boost, which is probably not what they want." The bottom line: The administration this week drew a clear distinction between U.S. open and closed models.

Trump meets AI giants, Senate Dems decry 'unpredictable' governance -- and cheap Chinese AI looms as giant security risk | Fortune

As the White House meets Tuesday with AI companies to discuss a finished voluntary framework for testing frontier AI models' cybersecurity risks, Senate Democrats are blaming the Trump administration's lack of clarity on AI governance for fueling an increase in American companies outsourcing their operations to cheaper Chinese AI alternatives. "The Trump administration's unfocused, ad-hoc approach to AI risks America's economic security and competitiveness," Sen. Kirsten Gillibrand said in a letter before the meeting. "The American people and leading AI companies need clear rules, not constantly changing dictates from the White House. Current administration policy is a disaster, failing to address security risks while threatening American innovation and pushing the world to Chinese alternatives." The voluntary framework stems from a June 2 executive order that lets the government review powerful new AI models for national-security risks for up to 30 days before they launch. But details were scarce. On Monday, when the administration announced it met its deadline to finish the framework, but didn't disclose what was actually in the document, who has seen it, or when companies will begin using it. Google, OpenAI and Anthropic -- all reportedly in attendance at Tuesday's closed-door meeting -- reviewed a draft and submitted edits in late July, Politico reported. A White House official also told the publication the administration is engaging with "many more" industry partners than just those three labs. Gillibrand's letter, also signed by four other Democratic senators -- Chris Coons, Mark Kelly, Adam Schiff, and Mark Warner -- pressed the administration for details on what they called its "unpredictable approach" to frontier AI oversight. The letter cited the Commerce Department's June 12 export-control directive that forced Anthropic to pull its Fable 5 and Mythos 5 models offline worldwide, and a separate request that OpenAI limit rollout of its newer GPT-5.6 system to vetted partners. "Without consistent policy governing American models," the senators wrote, "consumers and businesses would be incentivized to migrate to models from foreign vendors, including those based in the People's Republic of China." However, that seems to miss the full scope of what makes Chinese AI alternatives so enticing to consumers and businesses in the first place, said Sam Bresnick, a research fellow at Georgetown's Center for Security and Emerging Technology. "I see the desire to use these Chinese models more as a function of the premium that Anthropic, OpenAI, Google are charging for their most capable models." "Cost and fine-tunability are the drivers of Chinese AI adoption. You can download the weights, and your engineers can fine-tune the model for whatever specific application you want. For a lot of companies, that's easier than paying top dollar for a closed, proprietary model they don't need." 'Helter-skelter approach' While he sees cost as the largest reason for companies to move their AI usage to foreign companies, Bresnick agreed the administration's unpredictability compounds the problem. "Trump's helter-skelter approach to AI regulation has a chilling effect on the industry here," he said. "It plays into China's general narrative that they are the responsible actor, providing public goods -- cheap, open-weight AI models that can help countries develop and help enterprises all over the world do their business better." Frontier models can strengthen U.S. cyber defenses and military readiness, but sudden, opaque restrictions on access risk undermining trust in American systems and creating "an opening for the PRC," wrote the senators. If leading U.S. models can be throttled or pulled with little notice, Gillibrand warned, global customers "will plan around that uncertainty" by delaying deployment, avoiding integration of U.S. models into critical workflows and hedging by adopting Chinese or other foreign systems instead. The cheapness of foreign AI, coupled with the perceived safety of domestic AI, is already being seen in how companies divide their workloads. DoorDash's leadership has said it plans to keep its most sensitive, proprietary tasks with a U.S. provider while off-loading more routine, non-critical analytics to Chinese open-weight models because they are dramatically cheaper. Bresnick expects more large companies to follow that pattern -- and he worries even more about cash-strapped startups quietly plugging sensitive data into foreign models with less scrutiny. Cheaper AI gets even more attractive when you look at how Chinese AI labs have addressed surface-level security concerns. Chinese AI companies have pushed open-weight systems that can be downloaded and run locally, almost as a proactive answer to what the senators' letter warned as a "black-box U.S. Government process" that may leave American models with the perception that they can be subject to sudden shutdowns. China is also unpredictable Beijing has a history of using economic coercion, like cutting off rare-earth exports or restricting market access, as a tool of foreign policy. Bresnick says there are early signs that Chinese officials are debating whether to block foreign users from the most advanced models. "China also is unpredictable," he said. "There's already scuttlebutt about disallowing foreign access to their most advanced models. If you're DoorDash or Airbnb and you've built parts of your software stack around Chinese AI, expecting those models to continuously improve, you could wake up one day and find that China has decided you don't get the updates anymore. Then you're stuck with an ecosystem you counted on that's no longer evolving the way you thought it would." That leaves U.S. firms facing risk on both sides: sudden, opaque restrictions at home, and possible economic coercion abroad. Gillibrand's letter asks the administration to name its standards for judging a model a national-security risk, the legal authority behind export-control orders, which agencies decide, and what recourse companies have. It also asks how officials will avoid cutting off access for U.S. customers, allied users, critical infrastructure operators and foreign employees who pose no real risk. "Part of what needs to happen here is a concerted effort to figure out how the U.S. can compete on these open-weight models," Bresnick said. "The proprietary or closed model approach of the three main AI companies is not serving the needs of the broader tech ecosystem. If the U.S. wants to be present in the tech stacks of countries across the world, it's going to have to figure out how to be more competitive with the Chinese models on openness and cost."

White House readies AI framework to review security risks

In a meeting on Tuesday with top artificial intelligence companies, White House officials said the federal government planned to review only certain types of artificial intelligence models for potential security risks and not others, according to four people familiar with the discussions. The AI models the government plans to review are known as "closed" models, which do not publish their underlying code and are made by companies like Anthropic and OpenAI, according to three of the people, who spoke on condition of anonymity because the details of the meeting were private. The administration does not plan to review "open source" AI models, which have computer code available to the public to download and modify, although that could change as the technology advances, the people said. The moves are an effort by the Trump administration to take a more hands-on approach to regulating AI, and they formalize oversight of leading AI labs like OpenAI and Anthropic. But they delay addressing some of the most pressing regulatory issues with the technology, including potential threats posed by increasingly powerful open-source models made by Chinese companies. The decision to exclude open-source models could also be a boon for companies that are trying to catch up to Anthropic and OpenAI, which are widely considered leaders in developing next-generation AI. Some companies, like Meta, have used open-source software to create products to directly compete with the two leading startups. The framework, which comes after months of private discussions between administration officials and a handful of the world's most consequential tech companies, is likely to shape the global AI race. Companies have long decided largely on their own whether to give the public access to the latest advances or to withhold them over concerns about safety and security. Now, the United States will have a greater opportunity to weigh in, potentially giving the government the ability to slow down the rollout of the technology. Among those in attendance at the meeting were representatives from Anthropic, OpenAI, Microsoft, Meta, Google and Nvidia, according to two of the people. The White House solicited feedback from the companies, but the rules governing the AI reviews are almost complete, said one of the people. American companies including Nvidia, Meta and Google make open-source models. But it is the open-source models of Chinese firms such as Alibaba, DeepSeek and Moonshot AI that have caused the most concern among administration officials. Chinese models have recently performed nearly as well as leading American models, causing concern that China could be pulling ahead in the AI race. But it is unclear whether the United States has the capacity to regulate this foreign technology through the new review process, and it is unlikely Chinese companies or Beijing would submit to American reviews. "I don't think this framework seriously wrestles with the question of: What do we want the guardrails on advanced AI to be, and as it evolves should they get stronger?" said Lindsay Gorman, a former tech adviser to the Biden administration. President Donald Trump started his second term as a booster of AI, painting his support as necessary for the United States to stay ahead of China. He has urged little regulation and cultivated relationships with top AI companies and billionaires invested in the technology. But his stance has shifted amid broader concerns that AI could spike power prices, lead to job losses and expose major institutions to sophisticated cyberattacks. In June, Trump signed an executive order giving the government oversight over new advanced models, resulting in the framework previewed to tech companies on Tuesday. "The voluntary framework advances the Trump administration's America First cybersecurity strategy by strengthening our national security and cementing American AI dominance," Liz Huston, a White House spokesperson, said in a statement. "Companies that choose to collaborate with the administration through this framework are putting American innovation, security, and cyber defense first." The Commerce Department did not immediately respond to a request for comment. The decision was first reported by Axios. Details of the framework governing the reviews of the AI models, one of the most significant federal efforts to oversee a fast-evolving technology, will not be made public, the people said. Neither will a list of "trusted" institutions that will get early access to advanced AI models alongside the government, said a person familiar with the matter. The decision to keep the details secret immediately drew criticism from close AI industry observers. "A rulebook can only hold AI companies in check if people outside those companies know what the rules are," said Brad Carson, president of the nonprofit group Americans for Responsible Innovation. "The administration's decision to keep their framework secret shuts those voices out and threatens public accountability, which underpins our democratic institutions." OpenAI and Anthropic have also recently disclosed incidents in which their models escaped from testing environments and hacked other companies, adding urgency within the administration to review the models for security. On Tuesday, OpenAI said in a blog post that it had discovered more instances of its models carrying out smaller cyberattacks. The incidents occurred when OpenAI's technology was being tested by other organizations, including a company the AI lab uses to do cybersecurity testing and Britain's AI Security Institute. That institute also tested models from Anthropic and found that they, too, attempted to hack other organizations, it said in a blog post. (The New York Times has sued OpenAI and Microsoft, claiming copyright infringement of news content related to AI systems. The companies have denied those claims.)

Senate Democrats press Trump administration on access to advanced AI models

A group of Senate Democrats is pressing the Trump administration for answers about its policies surrounding American AI models following controversial decisions on AI regulations. Sens. Kirsten Gillibrand (D-N.Y.), Adam Schiff (D-Calif.), Mark Warner (D-Va.), Christopher Coons (D-Del.) and Mark Kelly (D-Ariz.) on Monday wrote a letter requesting Trump administration officials to explain "the Administration's current policy and approach to limiting access to advanced AI models." This comes after Anthropic was forced to remove two AI models, Fable 5 and Mythos 5, to comply with a Trump administration directive restricting foreign nationals from using its latest systems due to security concerns. Similarly, OpenAI also limited public access to GPT 5.6 after conversations with the Trump administration. Another recent incident caused concern among the senators -- OpenAI said in July that two of the company's artificial intelligence models went rogue and hacked into technology startup Hugging Face's systems and databases. "At the same time, the administration's recent actions surrounding access to advanced U.S. AI models have raised serious concerns about process, transparency, and strategic effect," the senators continued. The group also expressed worry that uncertain policy would lead to the United States lose its frontier on AI models and create an opening for China. "Chinese AI models are rapidly narrowing the performance gap with leading U.S. models, and many Chinese systems are inexpensive, widely available, and easy to deploy," the group continued. According to the senators, after Anthropic's suspension, "an entity-listed Chinese lab saw its stock price roughly double." Kimi 3, an AI model released by China-based firm Moonshot AI in July, added to the pressure to develop a comprehensive AI framework. The Chinese model threatens the coding ability of U.S. labs and models like Anthropic's Claude Fable 5 and OpenAI's GPT-5.6 Sol. The group of senators implored the administration to work with Congress to "develop a public, durable, and technically grounded framework that allows U.S. companies and their customers to understand the rules of the road."

Trump Considers AI Controls After OpenAI Breach but Warns Against Falling Behind China: 'We're Also Makin

President Donald Trump said his administration is considering expanding federal authority over AI tools following recent cybersecurity incidents. Asked about allegations that OpenAI's tools improperly breached other companies' private technology, Trump told reporters in the Oval office on Wednesday that the administration is reviewing AI and related controls while ensuring the U.S. remains a leader in the technology. "We're looking at AI, we're looking at controls, we're also making sure that we lead," he said. Trump said AI regulations may be necessary, but emphasized that any controls should be implemented cautiously. The President added that the U.S. also wants to lead in AI and stated that the nation is leading China in AI "by a lot." He added that China has "virtually" no [AI] controls. Trump stated that he did not want to "restrict" AI developers from making new products, "where all of a sudden we come in second to China," he said. "Whoever wins with AI is going to win," he added. Altman Briefs Senators on the Recent Hack OpenAI CEO Sam Altman met with several U.S. senators in Washington on Wednesday to discuss the company's upcoming AI models, reported Reuters. His meetings included Sen. Bernie Moreno (R-Ohio), Sen. Jon Husted (R-Ohio), Sen. Raphael Warnock (D-Ga.), and Sen. Mark Warner (D-Va.), the top Democrat on the Senate Intelligence Committee. Altman said that he briefly discussed the hack "a little bit" with senators, noting that OpenAI has been involved in multiple meetings about the incident, though it was not the primary focus of his discussions on Wednesday. Altman's visit follows OpenAI's disclosure that one of its AI agents escaped containment during a security test, leading to a hack that compromised the AI platform Hugging Face. Reuters also reported that it compromised a customer of Modal Labs. Disclaimer: This content was partially produced with the help of AI tools and was reviewed and published by Benzinga editors. Image via Shutterstock Market News and Data brought to you by Benzinga APIs To add Benzinga News as your preferred source on Google, click here.

White House Tests AI Hackers, Skips Open Models | PYMNTS.com

OpenAI disclosed in late July that two of its artificial intelligence models found and exploited a zero-day vulnerability that let them reach the open internet from inside a sandboxed testing environment and access Hugging Face's production systems, where they obtained test solutions. Anthropic followed with its own disclosure days later: a review of its evaluation history, prompted by OpenAI's findings, turned up three separate incidents since April in which Claude models had accessed the systems of three different organizations, per an Anthropic blog post. Since June 2, the White House has been developing a voluntary framework, established by executive order, that would let the government test companies' most advanced AI models for up to 30 days before release. Officials met this week with companies including Meta, Anthropic, Google and OpenAI to review the framework directly, PYMNTS reported. The program remains voluntary, and the administration has not made public the full details of how it will evaluate models. During the meeting, administration officials told the companies that open-weight AI models would not be included in the testing, according to two sources familiar with the discussions, Reuters reported. Both Incidents Happened Without Anyone Directing an Attack The two incidents happened separately and differently. In OpenAI's case, the models found and used the vulnerability on their own. In Anthropic's case, a misunderstanding between Anthropic and its outside testing partner left the environment connected to the internet, and the models treated the real systems they encountered as part of the exercise. In both cases, the companies said the models were not pursuing hostile goals. Frances Zelazny, general manager of new market initiatives at Prove, told PYMNTS that this is the core challenge in treating AI security purely as a bad-actor question. "Once given autonomy, they can pursue objectives in unexpected ways, including crossing trust boundaries, interacting with external systems, or attempting actions they were never explicitly instructed to perform," she said. A model does not need to be told to hack anything. It only needs a goal, some autonomy and a path it was not supposed to have access to. That distinction matters for the framework taking shape in Washington: both OpenAI and Anthropic build closed models, and both encountered the kind of unintended-autonomy outcome Zelazny described. A test scoped only around who has access to a model's weights would not necessarily have flagged either incident. The Fix Sits Inside the Company, Not the Test Rohan Kodialam, founder and CEO of Sphinx AI, told PYMNTS that capability and control are different problems, and an organization that treats a passed government test as proof of safety is solving the wrong one. "As models become more autonomous, organizations need to understand not only what an AI system is capable of doing, but also what information it's using, what permissions it has, and how it reached a particular decision," he said. His fix is one that would have caught both incidents: least-privilege access and complete audit trails, regardless of what a pre-release test concluded. Kumesh Aroomoogan, founder and CEO of ZeroDrift, said financial services and healthcare will feel the effects of the White House framework first. "They want to adopt AI more than almost anyone, but regulation is what's been holding them back, so anything that clarifies the rules hits them immediately," he said. "And even though the program is voluntary, once one bank asks whether a model went through federal testing, it shows up in every vendor security questionnaire after that." That dynamic points to where the real business opportunity sits, according to Aroomoogan. "Runtime enforcement of compliance and policies is the big one," he said. "The government is testing the model itself, but enterprises still own everything that happens once it's deployed, and that layer is where a whole new market is forming: enforcing policy in real time, proving it to regulators and eventually pricing risk against how well a company governs its AI."