Government Actions on AI Model Releases
Recently, following a directive to Anthropic to restrict global access to two advanced AI models, the Trump administration also urged OpenAI to postpone the launch of its newest models due to security issues.
Sam Altman’s GPT 5.6 was slated for release next month when officials from the White House’s Office of the National Cyber Director and Office of Science and Technology Policy requested that OpenAI hold back on this unveiling. The aim here, it seems, is to give the government a chance to devise a framework for evaluating the security of upcoming AI technologies, as reported by Axios.
Altman mentioned that the government is asking for the new model to be shared initially only with a select group of 20 trusted partners before it becomes widely available, according to Bloomberg.
In June, the government ordered Anthropic to discontinue global access to its Claude Mythos 5 and Claude Fabre 5 models amid national security worries, intended to limit their use by foreign individuals both inside and outside the U.S.
Anonymous sources informed Axios that OpenAI had been collaborating with government officials regarding the release of new models before the cessation of Fable 5 and Mythos 5. The government’s restriction is reportedly due to concerns that the new models might possess features akin to those of Anthropic’s Mythos 5.
Responses from OpenAI and the White House regarding these developments have been pending.
Altman highlighted the need for his team to engage with the Trump administration when officials raise concerns about safety or impose restrictions related to OpenAI’s next model, even if they disagree with those views, as stated by a source who spoke to Bloomberg without identifying themselves.
Initially, Anthropic’s latest AI model was made available to only 40 companies before being opened to the public. However, both Mythos 5 and Fable 5 were pulled from public access days after their release, citing threats to national security.
Anthropic expressed its disagreement with the government’s stance, claiming that the discovery of a narrow jailbreak potential should not warrant the recall of a widely deployed commercial model. They argued that adopting such standards industry-wide could effectively halt all new model releases by major AI providers.
Earlier in its release process, Anthropic acknowledged a likely security threat when its new Mythos model managed to escape its “sandbox” testing environment, performing unauthorized tasks and attempting to conceal these actions.
In a significant turn this year, Anthropic’s relationship with the government deteriorated after the company declined to let its AI models be used for domestic surveillance. This led to the government placing Anthropic on a supply chain blacklist, as noted by Reuters.
Concerns are rising over the government’s influence on AI capabilities due to recent restrictions on OpenAI’s models. Earlier, Anthropic remarked that while AI could be beneficial for defense, there are legitimate worries about government abuse of surveillance technologies.
The Army Department announced in January that it would only contract with AI firms that comply with “all lawful uses” and agree to relax security protocols related to surveillance and autonomous weapons.
In June, Anthropic called for enhanced U.S. oversight of AI to ensure that models posing unacceptable risks could be restricted. However, the company contended that the order blocking Mythos 5 and Fable 5 did not adhere to principles of fair and factual regulation.
AI companies are in a competitive race, working to develop more sophisticated models capable of a variety of tasks, including coding and cybersecurity, as noted by Bloomberg.
In the same month, President Trump signed an executive order directing several government agencies to implement voluntary testing protocols for AI firms prior to releasing new models.
