White House Finalizes New AI Safety Framework with Pre‑Deployment Model Reviews

On August 4, the White House finalized a new federal framework for overseeing advanced AI systems, according to multiple reports. The policy implements a June executive order and formalizes a pre‑deployment review scheme for so‑called “frontier models” – the most powerful large‑scale AI systems.

Under the new framework, leading labs such as OpenAI, Google DeepMind and Anthropic are expected to grant the U.S. government voluntary access to upcoming models for up to 30 days before public release. Companies would share safety test results, red‑teaming findings and broader societal risk assessments so that national‑security and critical‑infrastructure impacts can be evaluated in advance. While participation is technically optional, observers suggest the program could quickly become a de‑facto standard for major providers.

Analysts note that the approach resembles China’s 2023 rules requiring pre‑release security reviews and registration for generative AI services, raising questions about how far a democratic government should go in scrutinizing commercial AI. At the same time, firms are wary of exposing sensitive technical details or business secrets. Striking the right balance between transparency, innovation and geopolitical competition will likely define the next phase of U.S. AI governance.