A new Executive Order establishes mandatory pre-deployment safety testing standards for AI systems used in federal benefit determination, law enforcement, and national security applications.
President Biden signed Executive Order 14147 on April 14, 2026, establishing mandatory pre-deployment safety testing standards for artificial intelligence systems used in federal benefit determination, law enforcement decision support, and national security applications.
The order requires agencies deploying AI in these three categories to complete a documented safety evaluation including adversarial robustness testing, demographic disparity analysis across protected characteristics, and human-in-the-loop override documentation before any system goes live.
The order establishes a new AI Safety Testing Board within NIST, composed of technical experts from federal agencies, academia, and civil society, with authority to publish binding testing methodologies and reject inadequate agency safety evaluations.
For law enforcement applications specifically, the order requires that any AI system used in suspect identification, risk scoring, or sentencing support undergo independent bias auditing by an accredited third party with results published in machine-readable format.
The order directly expands the scope and urgency of GovAcademy's AI Ethics in Public Procurement course (GA-007) and AI Risk Register for Agencies course (GA-040), both of which cover the evaluation methodologies now mandated by the Executive Order.
GovAcademy's AI-Assisted Case Management course (GA-021) has been updated to reflect the new human-in-the-loop documentation requirements, including template AI Oversight Records and override procedures that agencies must maintain for EO 14147 compliance.
The American Civil Liberties Union and several technology advocacy groups praised the order as the most significant federal AI accountability action since the AI Executive Order of 2023.