OpenAI and Anthropic are each operating under an informal White House vetting process that limits who can access their most advanced AI models, marking an unprecedented level of government involvement in commercial AI product releases.
OpenAI announced Friday that its new model, GPT-5.6 Sol, is available only to roughly 20 customers pre-approved by the Trump administration. The company characterized the arrangement as temporary, describing it as a step toward broader availability in the coming weeks. OpenAI stated publicly that it does not believe this kind of government access process should become a long-term default.
Anthropic’s situation is more complicated. The company pulled two models, Fable 5 and Mythos 5, shortly after launch earlier this month following a Trump directive restricting their use by foreign nationals. On Friday, the administration lifted restrictions on Mythos 5, permitting its redeployment to a limited group of cyber defenders and infrastructure providers. Fable 5 remains offline.
Capability Concerns Drove Initial Scrutiny
The administration’s focus on Anthropic stems in part from the company’s own disclosures. Anthropic warned earlier this year that Mythos demonstrated an unusual aptitude for identifying software vulnerabilities in ways that could be exploited by malicious actors. David Sacks, who co-leads Trump’s council of technology and science advisers, said on a recent podcast that Anthropic CEO Dario Amodei effectively described Mythos as a cyber weapon during an April visit to Washington, which elevated concern across the administration.
OpenAI has taken a different public posture on Sol, stating the model is better suited to helping users find and fix vulnerabilities than to conducting attacks, and that it does not cross the company’s internal risk threshold. The company nonetheless acknowledged uncertainty about how Sol might behave when combined with other tools, which it cited as justification for the phased release.
Industry and Legal Pushback
Critics across the technology and security communities have raised concerns about the process itself. Stanford cybersecurity researcher Alex Stamos, who reviewed an Amazon analysis of Fable, said he found no risks beyond those present in other publicly available models, including models developed in China. Stamos argued that restricting domestic AI products while Chinese alternatives remain accessible undermines any stated goal of winning a global AI competition.
U.S. Representative Lori Trahan raised procedural objections, noting that the administration is making access decisions on a company-by-company basis without formal law, defined process, or legislative oversight.
Anthropic has had a particularly adversarial relationship with the current administration. The Pentagon previously designated the company a national security risk, and Trump ordered federal agencies to stop using Claude. Anthropic filed a lawsuit in response, which remains in federal court. The company said Friday it was pleased by the partial restoration of Mythos access and will continue working with the government to restore Fable’s availability.
Broader Policy Context
Trump signed an executive order in June establishing a framework for federal review of advanced AI systems for up to 30 days before public release. Participation by AI developers is described as voluntary, and the full framework has not yet been implemented. OpenAI CEO Sam Altman met with Commerce Secretary Howard Lutnick earlier this week as part of ongoing negotiations between AI executives and administration officials.