The proposal marks a shift in how federal regulators might approach oversight. Instead of rigid, static rules that risk becoming obsolete as technology evolves, the bill creates a licensing system for independent verification organizations. These entities would be held accountable for assessing whether AI models effectively mitigate catastrophic risks. If a verifier’s performance fails to meet public standards, the government retains the authority to revoke their license, forcing a market-driven approach to safety that keeps pace with technical breakthroughs.

Andrew Freedman, CEO of the nonprofit Fathom, argues that the current governance model is hitting a wall because legislation cannot outrun the speed of innovation. By fostering competition among verifiers, the act aims to ensure that accountability remains tethered to real-world outcomes rather than theoretical benchmarks. While Fathom supports the bill as a necessary starting point, the organization is pushing for further refinements, specifically advocating for federal tools that mandate risk mitigation during the early stages of model development rather than waiting until a crisis occurs.