The US government is urging Meta to submit its AI models for evaluation amid rising safety and security concerns about artificial intelligence technologies. The New York Times reported that Meta is the only major AI developer that has not voluntarily provided its models for review. The government aims to assess Meta’s capabilities and identify potential vulnerabilities in its AI technologies.
OpenAI and Anthropic are already collaborating with the government to test unreleased AI models according to Reuters. Meanwhile, companies including Google, xAI, and Microsoft have agreed to grant the Center for AI Standards and Innovation early access to their new models. The center, established by the Biden administration, is led by Commerce Secretary Howard Lutnick and staffed with experts to vet AI technology.
Meta spokesperson Francis Brennan stated, “We share the [Trump] administration’s goal of advancing US leadership on robust and secure frontier AI. While we are working through the details, we hope to sign the agreement soon.” The government has reportedly been communicating its requests to Meta via emails.
On June 2, President Trump signed an executive order establishing a framework for evaluating AI releases. Under this order, the government is required to create a review process by the end of July, aiming for companies to allow authorities up to 30 days to evaluate their technologies before public release. While an official review process is not yet in place, most major companies, excluding Meta, have been sharing their models with the government in advance.
Meta recently launched its latest model, Muse Spark, in April. This model features both “Instant” and “Thinking” modes, the latter of which provides reasoning capabilities by taking additional time to process prompts for comprehensive answers. However, Muse Spark is considered less powerful than other companies’ frontier models.
In mid-June, the government ordered Anthropic to suspend access to its Mythos 5 and Fable 5 models for foreign nationals due to national security concerns, prompting Anthropic to block access in compliance with the directive. Mythos, Anthropic’s leading cybersecurity AI model, has its latest version, Mythos 5, available only to Project Glasswing partners. Fable 5, designed to make many of Mythos’ capabilities accessible to the public, is noted to be less powerful but reportedly outperforms previous models.
