What happened
On 22 October 2026, Reps. Pat Harrigan (R-NC) and George Whitesides (D-CA) announced the AI Cyber Defense Act, shared exclusively with CyberScoop ahead of release. It would direct the National Institute of Standards and Technology (NIST) to develop voluntary industry standards and best practices for preventing 'distillation attacks' against US AI models by 'entities of concern' headquartered in Russia, China, Iran or North Korea (with NIST's director able to designate additional entities). NIST would first have to define what a distillation attack is, distinguishing it from legitimate open-weight/open-source sharing. The bill would also establish an 'AI Cybersecurity Roundtable' at NIST of government and AI-industry stakeholders to develop voluntary standards for defensive use of AI in critical-infrastructure cybersecurity. It follows the 16 October 2026 CISA/NSA/FBI joint advisory warning of industrial-scale state-backed distillation campaigns targeting OpenAI, Anthropic, Google and xAI models to bypass US export controls.
Why it matters
This is the first federal legislative attempt to treat knowledge distillation of US frontier models as a national-security and economic-espionage vector, and it directly responds to the White House/Commerce export-control posture that in June 2026 forced Anthropic to disable two models. If enacted, it would create NIST guidance that effectively defines what 'foreign' access to model outputs is permissible versus illegal, affecting every US model provider, cloud hyperscaler and enterprise API customer that distributes or consumes frontier-model output.
Action needed
AI developers and cloud platforms should monitor NIST standard development and begin documenting distillation-attack defenses (third-party access controls, output-token telemetry, account/geo verification) now, since voluntary standards can become the de facto baseline for both buyers and future enforcement. Track the bill number and committee referral once formally introduced.