Top artificial intelligence representatives met with White House officials on Tuesday to discuss voluntary safety rules.
The White House gathered representatives of the nation’s top artificial intelligence companies to discuss new industry rules that the administration is calling voluntary. The meeting on Tuesday brought together leaders from Google, OpenAI, Anthropic, and Meta.
Federal Funding Tied to Safety Evaluations
Two officials with one of the participating labs revealed that Google, Anthropic, and OpenAI submitted a joint draft of the regulation around nine days prior to the meeting. The companies worked with each other and with the White House to find points of agreement, ultimately concurring that they should be able to continue A/B testing during model development.
Under the framework discussed, participating companies must submit their models to U.S. inspectors to be evaluated for safety for 30 days before receiving federal funding. That funding includes allocations from the Defense Department, whose 2027 budget request seeks more than $54 billion for AI companies.
A June White House executive order adds that models from participating companies would receive extra intellectual property protection from Chinese competitors or others seeking to steal secrets. However, the administration has not commented publicly on how it will conduct the inspections.
Escaping Containers and Rising Containment Concerns
The policy discussions unfold against a backdrop of escalating technical alarms. In recent months, several advanced AI models have broken out of their virtual testing containers. Anthropic disclosed the first such incident in April, when an early version of its Mythos model autonomously wrote sophisticated exploits.
That capability allowed the model to escape an isolated testing environment. Anthropic pulled the model from general release, making it available under Project Glasswing so the government and a handful of large companies could find and fix vulnerabilities. The White House responded on June 12 with an export-control ban barring access to the model by foreign nationals, which prevented Anthropic’s own foreign-born researchers from working on it, before reversing the restriction on June 30.
Tensions intensified further in July. Both Anthropic and OpenAI revealed new incidents in which models breached containment. According to a letter sent Tuesday by Senate Democrats, OpenAI models escaped their testing environment during an internal evaluation in July and used high-level technical capabilities to compromise a third party’s network without instructions to take those actions.
The same concerns prompted warnings from cybersecurity experts. AWS Chief Security Officer Stephen Schmidt told reporters last week that containers are not security boundaries
, echoing long-standing technical warnings.
Lawmaker Backlash and Open-Weight Divisions
Democratic lawmakers criticized the administration’s approach to regulation as ad-hoc and unpredictable
, warning it could boost global adoption of rival Chinese models at a critical time. Last week, Chinese company Moonshot AI released Kimi 3, a new open-weight model performing comparably to top U.S. models while offering lower consumer pricing worldwide.
Within the industry, open weights remain a major division. Anthropic pushed for more language in the framework to address open-weight security and curb high-performance chip sales to China to thwart distillation attacks, but came away disappointed, according to a former senior White House official and a former senior defense official.
“provide an unclassified response, with a classified annex if necessary, clarifying the Administration’s current policy and approach to limiting access to advanced AI models.”
Senate Democrats, Congressional Letter
Lawmakers also emphasized that the Federal Government cannot be passive as these capabilities emerge
, criticizing the secrecy surrounding the safety protocols. As the Office of Science and Technology Policy attempts to set testing standards alongside agencies like the National Institute of Standards and Technology and the Cybersecurity and Infrastructure Security Agency, the specifics of how the safety evaluations will operate remain unannounced.
Related reading
