Bipartisan House Bill Would Task NIST With Voluntary AI Distillation Defense Guidance

A bipartisan House proposal, the AI Cyber Defense Act, would direct the National Institute of Standards and Technology (NIST) to develop voluntary standards and best practices for defending AI models against distillation attacks attributed to foreign entities. The bill is sponsored by Representatives Pat Harrigan, a Republican from North Carolina, and George Whitesides, a Democrat from California. It would also support convening government and industry stakeholders on defensive cybersecurity for critical infrastructure. The proposal remains legislation under consideration: it is not a solicitation, contract opportunity, or funding announcement, and it does not itself establish a new compliance requirement for agencies or contractors.

The bill’s focus is a growing concern for AI developers and national-security stakeholders: a distillation attack occurs when one model is trained using the outputs or reasoning of another, generally more capable model. NIST would first need to define the conduct the guidance is intended to address and distinguish harmful extraction from legitimate sharing and collaboration involving open-weight and open-source models. That distinction matters for both implementation and industry participation, because overly broad definitions could affect beneficial model development while narrow definitions could leave meaningful exposure unaddressed.

The source reports that the U.S. government and AI companies Anthropic, OpenAI, and Google have accused Chinese companies Moonshot AI, DeepSeek, and Alibaba of systematic efforts to distill and copy U.S. commercial models. Experts cited in the reporting describe alleged tactics involving large numbers of burner or resold accounts, false identities to bypass geographic restrictions, and thousands or even millions of prompts designed to extract useful capabilities. These are reported allegations, not findings established by the proposed bill. The legislation identifies foreign entities from Russia, China, Iran, and North Korea as a focus for the proposed voluntary guidance.

For GovCon participants, the near-term signal is policy and standards development rather than an immediate procurement. If the bill advances and NIST establishes the effort, AI security firms, model developers, critical-infrastructure cybersecurity providers, and federal contractors could be positioned to offer technical expertise through a proposed government-industry convening. Contractors should monitor legislative progress and consider whether their capabilities address model extraction, access controls, identity abuse, prompt monitoring, and protection of sensitive model outputs. NIST guidance would be voluntary under the proposal, so organizations should not treat it as a present agency requirement or assume that it will lead to a particular solicitation.

• The AI Cyber Defense Act would direct NIST to develop voluntary standards and best practices for defending AI models against distillation attacks.
• The bill is sponsored by Reps. Pat Harrigan (R-N.C.) and George Whitesides (D-Calif.) and is bipartisan.
• Proposed guidance would address threats involving foreign entities from Russia, China, Iran, and North Korea.
• NIST would need to define distillation attacks and distinguish malicious extraction from legitimate open-weight and open-source collaboration.
• Reported tactics include burner or resold accounts, false identities, geographic-restriction evasion, and high-volume prompting to extract model capabilities.
• The U.S. government and Anthropic, OpenAI, and Google have accused Moonshot AI, DeepSeek, and Alibaba of systematic model-distillation campaigns; these remain reported allegations.
• The proposal would convene government and industry stakeholders on AI cybersecurity for critical infrastructure, potentially creating a technical-engagement channel if established.
• AI security providers and federal contractors can evaluate capabilities in model access controls, prompt monitoring, identity verification, and protection of model outputs as potential contributions to the discussion.
• The bill does not itself provide funding, create a procurement requirement, or impose a compliance mandate; contractors should track its progress before making bid or compliance assumptions.
• George Whitesides said, “AI is advancing at a breakneck pace, and every month we keep China and other adversaries from stealing our progress is a month we gain to build the safeguards this technology needs.”