Not sure where to start? Answer 3 questions and get a tailored compliance action plan.
What applies to me? →Promoting Advanced Artificial Intelligence Innovation and Security
Issued by
United States White House
This presidential action establishes a classified benchmarking process to evaluate frontier AI models for advanced cyber capabilities and introduces the defined category of 'covered frontier models' for US AI governance purposes. It applies primarily to AI developers building or deploying large-scale frontier models and creates a voluntary pre-release access-sharing arrangement with the federal government. The action focuses on national security risk reduction, insider-threat mitigation, and structured government-industry collaboration on model safety evaluation.
Applies To
Overview
Issued by the White House on June 10, 2026, this presidential action directs the creation of a classified benchmarking regime to systematically assess whether covered frontier AI models possess dangerous or exploitable cyber capabilities before public release. The action formally defines 'covered frontier models,' establishing a threshold-based category that determines which AI systems fall within scope of associated security obligations. A voluntary framework is introduced under which AI developers may provide pre-release model access to designated government evaluators, enabling classified safety and security assessments prior to deployment. Insider-risk protection measures are incorporated to govern how federal personnel and contractors handle access to sensitive model weights and evaluation data. The action does not currently impose mandatory reporting or registration requirements on developers, but signals a trajectory toward formalized obligations as evaluation infrastructure matures. Enforcement mechanisms and interagency coordination responsibilities are to be defined through subsequent implementing guidance.
Key Requirements
- •Developers of 'covered frontier models' must understand whether their systems meet the defined threshold criteria triggering applicability under this action.
- •Participating developers who opt into the voluntary framework must provide pre-release model access to designated government evaluators under structured access protocols.
- •Insider-risk protection controls must be implemented by both developers and government personnel handling model weights or evaluation outputs under the program.
- •Classified benchmarking assessments conducted by government evaluators focus specifically on advanced cyber capabilities, requiring developers to cooperate with scoping and data requests.
- •Developers participating in the voluntary framework should maintain documentation of access grants, evaluation timelines, and any resulting government feedback for internal governance records.
- •Subsequent implementing guidance is anticipated and organizations must monitor for mandatory requirements that may follow this foundational presidential action.
What Your Organization Must Do
- →Audit all AI models in development or deployment to determine whether they meet the definition of 'covered frontier model' as specified in the action and any subsequent implementing guidance.
- →Establish an internal tracking process to monitor regulatory developments and implementing guidance that may convert voluntary obligations into mandatory requirements.
- →Engage legal counsel to evaluate the legal and contractual implications of voluntarily sharing pre-release model access with federal government evaluators.
- →Implement or review insider-risk protection policies covering access to frontier model weights and sensitive evaluation data to align with the action's security requirements.
- →Update AI governance documentation to reference this presidential action and assign ownership for monitoring follow-on rulemaking or agency-level directives.
- →Coordinate with government relations and policy teams to assess whether voluntary participation in the benchmarking framework offers strategic or reputational benefits that warrant early engagement.
Playbook Guidance
Step-by-step implementation guidance for compliance teams.
Frequently Asked Questions
- What is a 'covered frontier model' under the PAIAIS presidential action and how is the threshold determined?
- The action formally defines 'covered frontier models' as a threshold-based category, but precise numerical or computational criteria are to be clarified through subsequent implementing guidance. Developers should audit current and pipeline models against any published thresholds and monitor agency-level directives for updated scoping criteria as the regulatory infrastructure matures.
- Is participation in the PAIAIS pre-release government benchmarking program mandatory for AI developers?
- No, participation is currently voluntary. Developers may opt into a structured framework that provides pre-release model access to designated federal evaluators, but no mandatory sharing requirement exists yet. The action signals that formalized obligations may follow as evaluation infrastructure develops, so organizations should monitor implementing guidance closely.
- What are the insider-risk protection requirements under PAIAIS and who do they apply to?
- The action requires both participating developers and federal personnel handling model weights or evaluation outputs to implement insider-risk protection controls. Specific technical or organizational standards will be defined in subsequent guidance. Developers should review existing access-control and data-handling policies now to identify gaps before implementing rules are issued.
- What legal and contractual risks should counsel assess before voluntarily sharing pre-release model access with federal evaluators under PAIAIS?
- Counsel should evaluate IP protection, confidentiality obligations, liability exposure if evaluation findings influence deployment decisions, and any conflict with existing investor or licensing agreements. The classified nature of benchmarking assessments creates additional complexity around how feedback can be used or disclosed internally.
- Does PAIAIS impose any reporting or registration deadlines on AI developers before the June 10, 2026 effective date?
- No mandatory reporting or registration requirements are currently imposed. The action is foundational, with enforcement mechanisms and interagency coordination responsibilities deferred to implementing guidance. Developers should use the pre-effective period to assess model applicability and establish internal tracking processes for follow-on rulemaking.
- How does PAIAIS compare to the EU AI Act in terms of obligations for frontier AI developers?
- Unlike the EU AI Act, which imposes mandatory conformity assessments and market registration for high-risk systems, PAIAIS currently relies on voluntary industry participation and classified government-led evaluation focused specifically on cyber capabilities. PAIAIS is narrower in scope and US-specific, but may evolve toward mandatory requirements through subsequent presidential or agency directives.