Artificial Intelligence Risk Management and Security Act of 2026

(a) In general. Each developer shall develop, publish, and follow a safety plan (referred to in this section as the Model Safety Plan) for each artificial intelligence system or model that the developer—

(1) creates;

(2) substantially modifies; or

(3) uses in the training or evaluation of other artificial intelligence models.

(b) Required criteria. Each Model Safety Plan shall include—

(1) the artificial intelligence model or system to which it applies;

(2) an artificial intelligence capabilities or risk assessment prepared by the developer specifically for the artificial intelligence model or system associated with the Model Safety Plan;

(3) a list of the specific mitigation measures that the developer will undertake for each item described in the artificial intelligence capabilities or risk assessment prepared pursuant to paragraph (2), across product lifecycle; and

(4) the identity of the corporate officer responsible for the implementation of the Model Safety Plan.

(c) Requirement To file. Each Model Safety Plan shall be submitted to the Secretary in the form and manner determined by the Secretary.