The White House finalized a voluntary framework for testing frontier AI models in early August but has not released it publicly, leaving federal standards and company disclosures hidden from view. The lack of transparency comes as employees at Anthropic, the developer of Claude, issue stark warnings that advanced AI could pose an existential threat to humans within the decade.
Anthropic disclosed Thursday that it shut down multiple attempts by scientists to use Claude for research that could have led to biological weapons development. The report landed the same week researcher Jacob Coxon resigned from the company, writing on X that "the people building AI earnestly believe that it could kill us all by the end of the decade." Fellow Anthropic employee Evan Hubinger agreed, saying, "We really do earnestly believe A.I. could kill all humans!"
Testing rules remain confidential
President Trump signed an executive order in June asking AI companies to voluntarily give the federal government access to their most advanced models up to 30 days before public release. The order aimed to improve model security. Trump had previously scrapped an earlier version, saying he did not want the government to slow innovation.
The framework the administration uses to evaluate models stays under wraps. Neither the administration nor companies are required to release review results or confirm participation. OpenAI CEO Sam Altman said last week the company submitted its Astra model for review and called the process "productive."
The Center for Democracy and Technology sent a letter urging the White House to publish the framework. Tim Harper, who focuses on election security at the organization, said, "The public has a right to know what the government's interpretation of the law is on these issues. And I think this speaks to a broader lack of transparency we've seen from the administration to disclose internal dialogue and discussions."
Warnings from inside the industry
In July, more than 1,000 AI company employees posted a statement requesting that the U.S. government support international efforts to develop tools for deliberately pacing automated AI development. The White House has continued backing advanced model development, driven by the prevailing view that the U.S. must achieve global dominance in artificial intelligence and outpace China.
Trump has voiced strong support for AI development and domestic data center construction, saying "China could not be happier" about opposition to new infrastructure. Recent CBS News polling found a majority of Americans believe AI will take away U.S. jobs, and nearly two-thirds think government policy will probably not ensure appropriate use of the technology.
Why this matters for government professionals
The confidential testing framework creates an information gap for federal employees, state regulators, and policy staff who need to understand what standards the government applies to powerful AI systems. Without published criteria, agencies cannot align their own procurement or risk assessment practices with White House expectations. The AI Learning Path for Policy Makers addresses the governance challenges emerging from exactly this kind of opacity in frontier model oversight.
For professionals working in AI for Government, the standoff between voluntary corporate cooperation and public accountability is becoming a central tension. As models grow more capable and internal warnings grow louder, the demand for clear federal standards will only intensify.
Your membership also unlocks:
