Trump Administration Finalizes AI Pre-Review Framework; OpenAI, Anthropic, Google Submit Revised Drafts

The Trump administration is concluding efforts to establish a new self-regulatory framework requiring AI companies to submit models for federal review before public release. The Office of the National Cyber Director (ONCD) delivered a draft framework to OpenAI, Anthropic, and Google approximately two weeks ago, with the firms jointly submitting feedback. This framework, mandated by an executive order signed by President Trump in early June, aims to be finalized by August 1. The order establishes a 30-day federal review period for high-risk AI models to address cybersecurity vulnerabilities, with a focus on advanced systems capable of identifying security flaws. The initiative seeks to standardize procedures that previously included ad hoc measures like withholding public access to Anthropic's cybersecurity-focused model 'Miso' and imposing export controls on its advanced models. The new framework aims to replace temporary measures by defining criteria for classifying AI models as 'frontier models' and establishing standardized review processes before external deployment. Industry concerns remain, particularly over the definition of frontier models and whether open-source AI should be subject to the same regulations as closed-source models, with fears that excessive oversight could hinder the growth of the U.S. open-source ecosystem.