Washington Mandates Frontier AI Safety Audits

Washington State has enacted the Artificial Intelligence Accountability Act, becoming the first US jurisdiction to mandate annual third-party audits for certain frontier AI models. The legislation, signed into law by Governor Jay Inslee on 10 July 2026, aims to establish robust safety standards for powerful AI systems.

The Act targets models categorised as 'frontier AI models'. These are defined as general-purpose AI models requiring at least 10^26 floating point operations (FLOPs) for training. The law also includes models that exhibit capabilities deemed 'highly dangerous' or that could pose a 'catastrophic risk to national security, national economic security, or national public health and safety'. Such determinations will be made by the newly formed Washington State AI Office.

Covered entities, including developers and deployers of these frontier models, must submit to annual, independent third-party audits. These audits will assess several critical aspects: model safety, fairness, privacy, and security. The auditing process mandates a 'red-teaming' exercise, simulating adversarial attacks to identify vulnerabilities. Furthermore, an assessment of the model's capacity for autonomous replication and adaptation is required.

Organisations must also conduct and report on a 'catastrophic risk evaluation'. This evaluation will consider potential harms such as widespread disinformation, cyberattacks, chemical or biological threats, and scenarios where AI systems become uncontrollable. The Washington State AI Office will develop specific audit methodologies and criteria. They will also oversee the registration of approved third-party auditors.

Non-compliance with the Act carries significant penalties. Civil penalties can reach up to 1% of the organisation's global annual revenue for the preceding year, or a maximum of $100 million, whichever is greater. This enforcement mechanism underscores the state's commitment to ensuring accountability in the development and deployment of advanced AI.

The legislation represents a proactive step in state-level AI regulation. It seeks to balance innovation with public safety through external oversight. Will this state-level initiative inspire a broader federal or international regulatory framework for frontier AI?