President Donald Trump signed an executive order on June 2, 2026 [1], requiring voluntary pre-release safety reviews for advanced artificial intelligence models [1].
This move attempts to balance the mitigation of national security risks with the need to maintain U.S. technological leadership. By keeping the reviews voluntary, the administration aims to avoid rigid regulations that could stifle innovation or hinder the competitiveness of American firms against global rivals [2].
Under the new order, developers are asked to submit their models for review up to 30 days before a public launch [1]. The process is designed to identify potential misuses of powerful systems, such as Anthropic’s Claude-Mutes, before they reach the general population [3].
Government officials said the initiative relies on industry cooperation rather than mandatory enforcement [2]. This approach is intended to curb the risks associated with the misuse of high-capability AI, while allowing companies to maintain their development timelines [3].
While some reports suggested the administration was only considering such a move, the signing of the executive order on June 2, 2026 [1], establishes the formal framework for these voluntary submissions [1]. The administration believes that cooperation with the private sector is the most effective way to secure AI systems without imposing a heavy regulatory burden [2].
The order focuses specifically on advanced models that possess capabilities likely to impact national security [3]. By establishing a window for review, the U.S. government seeks to create a safety buffer that allows for the identification of critical vulnerabilities before a model is deployed globally [1].
“President Donald Trump signed an executive order on June 2, 2026, requiring voluntary pre-release safety reviews.”
The shift toward voluntary reviews indicates a preference for a 'partnership' model of governance over a 'regulatory' model. By avoiding mandatory mandates, the U.S. government is attempting to prevent a 'brain drain' or a slowdown in development that could allow other nations to achieve AI supremacy, while still creating a mechanism to flag catastrophic risks before they go live.
