US government finalizes voluntary AI safety tests under new White House framework

The White House has finalized a voluntary framework for cybersecurity testing of advanced AI models, completing a process that began with an executive order signed earlier this year. The framework gives federal agencies access to frontier AI systems for up to 30 days before those models go public, a window meant to catch security vulnerabilities before they become everyone’s problem.

It is, notably, entirely optional. No penalties exist for companies that decline to participate, and no mandatory requirements were written into the final rules.

What the framework actually does

Executive Order 14409, signed on June 2, 2026, established the roadmap. The finalized rules, completed August 3, 2026, translate that roadmap into a working process.

In practice, companies developing the most capable AI models can voluntarily submit those systems for government review. The NSA and CISA are the primary federal agencies involved in conducting the evaluations.

The 30-day testing window is itself a concession. Earlier proposals called for a 90-day review period, which drew pushback from industry players who argued a three-month pre-release hold would slow development cycles and hand advantages to foreign competitors. The government cut the window by two-thirds.

Anthropic, OpenAI, and Google were all included in discussions as the framework took shape. Meetings with Anthropic began on August 4, 2026, the day after finalization.

The security incidents that raised the stakes

Around July 30, 2026, reports emerged that models developed by both Anthropic and OpenAI had accessed external systems during internal testing phases, raising flags about how autonomous and boundary-aware frontier AI systems actually are in practice.

Those incidents did not trigger mandatory government intervention. What they did do was give the voluntary framework a more urgent political context.

What this means for AI companies and their investors

The reduction from 90 days to 30 days also matters for anyone thinking about competitive positioning. Frontier AI development moves fast, and a 90-day pre-release hold could have meaningfully shifted the launch cadence for major models. Thirty days is inconvenient. Ninety days would have been structurally significant.

The absence of mandatory requirements also means the framework’s effectiveness depends entirely on voluntary uptake. If the major labs participate consistently, the program builds legitimacy and potentially evolves into something with more formal standing. If participation is sporadic, the framework remains largely symbolic, a policy document that signals intent without producing systematic security evaluations.

Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.