Trump Administration is reportedly nearing a voluntary AI standards deal with major technology companies as Washington sharpens its focus on frontier model risks.
Key Points:
- The reported agreement would set standards for frontier AI models, with cybersecurity capabilities at the center.
- The Financial Times said CAISI and the NSA are expected to play central roles once the standards are formalized.
- The talks follow recent U.S. pressure on Anthropic and a shift away from the administration’s early light-touch stance.
Trump AI Standards
The Financial Times reported that the Trump administration could announce the standards “as early as next week,” citing people familiar with the talks.
The agreement would be voluntary and would involve several major U.S. frontier AI companies. It is expected to focus on model capabilities that could affect cybersecurity, now a central part of Washington’s AI risk agenda.
One person cited by the FT said the Center for AI Standards and Innovation, which sits under the Commerce Department, and the National Security Agency, which sits under the Pentagon, would help guide the standards after they are formalized.
The report did not make clear which companies would sign on. It mentioned Anthropic, OpenAI, Amazon, Microsoft and Google, but not Meta, which earlier reporting described as a holdout.
Also Read: Experts Say Open USD Could Do Something USDC Never Saw Coming
Anthropic Pressure
The expected deal follows a clear change in the administration’s approach to AI oversight.
At the start of the second Trump administration, J.D. Vance signaled a more hands-off approach to AI regulation. That stance has since narrowed, with White House action against Anthropic, a new executive order on AI and the reported standards talks.
On Jun. 12, the U.S. delivered an export control directive to Anthropic that disabled its latest publicly released model for the rest of June. OpenAI has reportedly held back its newest models, apparently to avoid a similar disruption.
The executive order calls for a classified benchmarking process to assess advanced cyber capabilities in AI models. It also directs officials to determine when a model should be treated as a “covered frontier model.” That structure would leave the public without full visibility into the tests used on frontier AI systems. Still, common safeguards adopted across several companies could reveal parts of the standards over time.
The talks build on a broader shift in Washington’s AI policy. A voluntary deal would not settle every regulatory question, but it would give companies a clearer process after weeks of uncertainty.
Read Next: Bitcoin Ownership Below 1% Fuels CZ’s Boldest Call Yet: $1M Target
