US senators propose an AI Safety Board with 45-day pre-release access to frontier model weights; Cruz blocks fast-track passage

Why it matters

【Key points】

Senators Mark Warner (D-VA), Brian Schatz (D-HI) and Andy Kim (D-NJ) have proposed the Artificial Intelligence Risk Management and Security Act of 2026 (S. 5576). It would create a permanent AI Safety Board in the Commerce Department, drawing on NIST, CISA, NSA, Treasury and independent experts, to set enforceable safety standards for frontier models. Developers would have to give the Board access at least 45 days before public release, including weights, configuration files, runtimes and libraries; follow Model Safety Plans naming risks, mitigations and a responsible officer; and face civil penalties of up to $250,000 per violation per day. Serious incidents would be reported within 30 days, or 72 hours when they threaten national security, critical infrastructure or public safety. On September 29 the three sought unanimous consent to pass it; Sen. Ted Cruz (R-TX) objected.

【Verifiable】

Per the congressional record, S. 5576 was introduced on September 29 and referred to the Senate Commerce Committee, where it remains; its only cosponsors are Schatz and Kim. Provisions come from Warner's office press release; Cruz's objection is in Warner's September 29 floor statement release. The bill also calls for a national AI incident database, secure federal testing using NSA and national lab resources, and standards and documentation for autonomous AI agents.

【Our read】

This is the toughest federal pre-release review proposal for frontier models so far: not self-assessment, but handing weights to government testers six weeks before launch, backed by daily fines. Near-term odds are poor: consent was blocked, the bill sits in a committee Cruz chairs, and it has no Republican cosponsor. Its real effect is to anchor the debate; figures like 45 days, weight access and 72-hour reporting are likely to resurface in bipartisan or state versions.