US Senate moves toward sweeping AI law forcing companies to curb major risks

A bipartisan group in the US Senate is converging on the outline of a federal AI safety bill that would require developers of powerful systems to “mitigate known major risks” before deployment. The talks mark a new phase in Washington’s response to growing worries over how frontier AI models and autonomous agents could be misused for cyberattacks, national security threats, or large‑scale social harm.

According to recent reporting, the draft framework would obligate companies offering advanced AI systems to conduct risk assessments from the design stage, implement concrete safeguards, and report both their methods and residual risks to federal authorities. Firms that ignore serious, well‑documented dangers could face penalties. At the same time, negotiators are wrestling with fears that heavy‑handed rules could choke off innovation at smaller startups, so one option on the table is to focus obligations on “frontier” models at the very top end of capability.

This bill is viewed as one of the few AI measures that could realistically pass Congress in 2026, and its reach would extend far beyond US‑headquartered vendors, touching any company that offers high‑risk AI services in the American market. Alongside the EU’s AI Act and China’s algorithm rules, it is likely to become a key reference point for how governments around the world choose to govern advanced AI.