Eigen RadarAI
Analysis

US lawmakers propose outside reviews for advanced AI models

US House members Don Beyer, Sara Jacobs and Celeste Maloy have released an outline for federal rules on developers of the most advanced artificial intelligence models. Their proposal calls for companies to publish safety plans, report certain incidents and submit models for independent evaluation. The federal rules would set a minimum standard and leave states free to impose additional safeguards.

Artificial Intelligence··Midday
Don Beyer, one of the three House sponsors of the AI safety outline, walks a corridor of a House office building in Washington.

The outline calls for disclosures and incident reports

On 19 November, US representatives Don Beyer, Sara Jacobs and Celeste Maloy released an outline of proposed federal safety rules for developers of the most advanced artificial intelligence models. It would require covered companies to publish plans explaining how they manage safety risks and to report certain incidents to federal authorities.[1], [2]

Developers would submit models for independent evaluation

The outline calls for outside evaluation before commercial release or widespread use within the company. It proposes a role for a federal standards and evaluations centre and for independent organisations accredited by the government. It would bar companies from paying the organisations that evaluate their models directly and set conflict-of-interest rules for employees moving between developers and reviewers.[1]

States could keep their own additional rules

The proposal would set a federal minimum while letting states enforce extra safeguards. California already requires covered developers to publish safety plans and report certain serious incidents. The proposal generally envisages leaving startups and academic researchers outside the federal requirements. As they develop legislative text, the sponsors are consulting experts and working with the relevant committees.[1]

References

  1. News sourceCASRAIUS House lawmakers outline safety requirements for frontier AI developers↩1↩2↩3
  2. News sourceOffice of U.S. Representative Don BeyerUS House members outline federal safeguards for advanced artificial intelligence↩