White House Keeps AI Safety Framework Secret Amidst Growing Fears
The administration has finalized a voluntary testing regime for advanced AI models but refuses to make the rules public, leaving experts and the general public in the dark about critical safety standards.
The White House has completed a voluntary framework for testing frontier artificial intelligence models, but the document remains confidential. There is no public schedule for its release, creating a significant gap in oversight just as warnings about AI safety become more urgent.
This silence persists despite recent reports that Anthropic, the developer of Claude, blocked multiple attempts by researchers to use the system for biological weapon research. The lack of transparency means neither the government nor the private sector is required to disclose what standards are being met or if new breakthroughs are being flagged.
Confidential Reviews Hide Safety Standards
According to reporting by GN technics/ai (en-US), the framework was finalized in August following an executive order signed by President Trump in June. The order asked major AI firms to grant federal access to their most advanced models up to 30 days before public release. However, the specific criteria used for these evaluations are not public.
OpenAI CEO Sam Altman recently described the process as productive, noting that his company submitted its new Astra model for review. He emphasized the need for close engagement with safety institutes in the U.S. and abroad as capabilities grow. Yet, without public benchmarks, it is impossible for outsiders to verify if these reviews are rigorous or merely procedural.
Industry Insiders Warn of Existential Risk
Concerns within the AI sector are intensifying. Jacob Coxon, a researcher at Anthropic and former OpenAI employee, recently resigned, stating on social media that he believes the technology could kill humans by the end of the decade. Evan Hubinger, another Anthropic employee, echoed this sentiment, asserting a sincere belief in the potential for catastrophic harm.
These warnings align with a broader trend. In July, over 1,000 employees from various AI companies signed a statement urging the government to support international efforts to pace the development of automated AI. They argued that deliberate governance tools are necessary to manage the rapid advancement of the technology.
Political Pressures Favor Speed Over Caution
Despite internal and external warnings, the administration continues to prioritize rapid development. The prevailing belief is that the United States must maintain global dominance in artificial intelligence and outpace China. President Trump has expressed strong support for AI infrastructure, arguing that opposition to new data centers would please Chinese competitors.
This approach creates a distinct trade-off. Speed and competitive advantage are prioritized over transparent safety verification. A majority of Americans believe AI will take jobs, and nearly two-thirds doubt that current government policies will ensure the technology is used appropriately. The absence of a public framework makes it difficult for the public to hold the administration accountable for these risks.
