US government reportedly urges Meta to share its AI models for safety review

At a glance:

• Meta is the only major AI developer not voluntarily sharing models for government evaluation amid rising safety concerns.
• OpenAI, Anthropic, Google, xAI, and Microsoft have already agreed to provide early access to the Center for AI Standards and Innovation.
• President Trump's June 2 executive order mandates a federal review process for AI releases, with companies expected to allow 30-day evaluations before public launches.

What the US government is asking of Meta

The US government is reportedly pressuring Meta to submit its AI models for evaluation through the Center for AI Standards and Innovation, a Biden-era agency staffed with technical experts to vet AI technology. According to The New York Times, Meta remains the sole major AI developer that has not voluntarily turned over its models for review, despite growing concerns about the risks posed by advanced artificial intelligence systems. The agency, headed by Commerce Secretary Howard Lutnick, aims to assess capabilities and identify vulnerabilities in frontier AI models. Meta spokesperson Francis Brennan acknowledged the Trump administration's goals, stating, "We share the [Trump] administration's goal of advancing US leadership on robust and secure frontier AI," while noting ongoing negotiations to finalize an agreement.

Other companies already cooperating with federal oversight

While Meta has yet to comply, several competitors have already begun sharing their models with the Center for AI Standards and Innovation. OpenAI and Anthropic are actively working with the government to test unreleased models, according to Reuters. Google, xAI, and Microsoft have also committed to providing early access to their new AI systems. These collaborations come as the federal government seeks to establish a standardized review process for AI technologies before public release. The initiative reflects a broader push to balance innovation with national security and ethical safeguards, particularly for models with advanced reasoning or cybersecurity capabilities.

Trump's executive order and the race to establish AI review protocols

President Trump signed an executive order on June 2 establishing a framework for federal AI evaluation, requiring the government to develop a review process by the end of July. Under the order, companies are expected to grant authorities up to 30 days to assess their technologies before public launch. However, the framework is still in its early stages, and no official review process has been formalized yet. Despite this, the aforementioned companies have reportedly been sharing their models voluntarily for months, signaling proactive engagement with regulatory efforts. The order underscores the administration's focus on maintaining US leadership in AI while addressing potential risks tied to rapid technological advancement.

Meta's Muse Spark model under scrutiny

Meta launched its latest AI model, Muse Spark, in April, featuring two distinct modes: "Instant" for rapid responses and "Thinking" for enhanced reasoning capabilities. The Thinking mode allows the model to take additional time to process prompts, delivering more thorough answers. While Meta describes Muse Spark as less powerful than other frontier models, the government's heightened scrutiny of the AI industry has placed even mid-tier models under closer examination. The model's dual-mode functionality highlights Meta's efforts to balance performance with user experience, though its reluctance to share models for review has drawn criticism from regulators and competitors alike.

Anthropic's compliance with national security directives

In mid-June, the US government ordered Anthropic to suspend access to its Mythos 5 and Fable 5 models for all foreign nationals, citing national security concerns. Anthropic fully blocked access to both models to ensure compliance, even though Mythos 5 is its state-of-the-art cybersecurity AI model, available exclusively to Project Glasswing partners. Fable 5, designed to bring Mythos' capabilities to the public, was noted by Anthropic as outperforming any previous model it had launched. The directive highlights the government's willingness to impose restrictions on AI technologies deemed sensitive, even as it encourages voluntary cooperation from developers.

Implications for AI regulation and industry dynamics

The push for Meta to share its models reflects a growing tension between AI innovation and regulatory oversight. As the industry races to deploy frontier models, governments worldwide are grappling with how to manage risks without stifling progress. For Meta, the pressure to comply could signal a shift toward stricter compliance standards, particularly for models with dual-use potential. Meanwhile, the cooperation of other major players suggests a willingness to engage with federal frameworks, potentially setting a precedent for future AI governance. The outcome of these negotiations may influence how AI regulation evolves in the US and beyond.