The White House is in advanced talks with OpenAI, Google, and Anthropic to finalise voluntary standards for frontier AI model releases, the Financial Times confirmed, with an announcement possible as soon as next week.
Technical teams from the three AI labs have met consistently with White House officials to negotiate two sticking points: how long the government can hold a model before public release, and what capability threshold triggers a review. The framework would replace what former White House AI adviser Dean Ball called a "de facto involuntary licensing regime" — the ad-hoc system that pulled Anthropic's Fable 5 offline for 20 days and intercepted OpenAI's GPT-5.6 before launch.
The operative variable is cybersecurity benchmarks. GPT-5.6 Sol scored 96.7% on OpenAI's internal Capture the Flag evaluation, crossing the "High" risk threshold. Google's Gemini 3.5 Pro scored 70.7% on TerminalBench 2.1 — over 18 points lower — which is why Gemini escaped restrictions entirely.
Analysis
A formal voluntary framework is a regulatory win for labs heading to public markets. Both Anthropic and OpenAI list the current ad-hoc review regime as a material risk in their S-1 filings. Predictable timelines and defined rules shrink that risk before the IPO roadshow window opens. The real test is whether "voluntary" standards stay voluntary when the next Fable-level model ships.
