The Trump administration has requested Meta to voluntarily submit its AI models for federal review to assess their capabilities and vulnerabilities, according to multiple sources [1, 2, 3]. Meta is currently the only major US AI developer that has not agreed to such government-led voluntary reviews [1, 2, 3]. A confidential email from the administration pressed Meta on June 23 to participate in the process, emphasizing national security concerns [1, 2]. Meta launched its Muse Spark AI model, which features "Instant" and "Thinking" modes, in April 2026 [1, 3]. Meta spokesperson Francis Brennan said the company shares the administration's goal of advancing US leadership in frontier AI and is “working through the details” to sign an agreement soon [3].
OpenAI on June 25-26 announced a limited rollout of its GPT-5.6 models, providing access to about 20 government-approved trusted partners ahead of a wider release, in line with a Trump administration request citing security risks and national concerns [4, 5, 6, 7]. The government requested up to 30 days of review before public releases of AI models, established by an executive order signed by President Donald Trump on June 2 [1, 2, 3]. OpenAI said the limitations protect against higher-risk activities including sensitive cybersecurity tasks, with its GPT-5.6 Sol model targeting coding, biology, and security applications [5, 6].
OpenAI CEO Sam Altman called the government’s limited access process “bad news” and “not our preferred long-term model,” adding the company is working with government and industry toward a “more sustainable approach” but supports the short-term step to enable broader availability [5, 7]. An unnamed US source noted the administration wants companies to have “adequate safeguards in place” due to the high power of these models [7].
Anthropic, another leading AI company, was ordered in mid-June to suspend foreign access to its advanced models over national security worries [1, 5, 6, 2, 3]. Among major AI labs, OpenAI, Anthropic, Google, xAI, and Microsoft have all agreed to voluntarily provide unreleased models for federal review [1, 2, 3]. Commerce Secretary Howard Lutnick is involved in ensuring models undergo government testing and approval [7, 3].
The US government aims to identify AI risks such as cyberattacks and military misuse before frontier models see full deployment [1]. The administration has set a deadline to establish a formal AI model review process by the end of July 2026 [2, 3]. Meta’s response and the ongoing limited release of GPT-5.6 mark critical developments in federal oversight of powerful AI technology.
