Bipartisan bill tasks NIST with creating AI cyber defense standards

A bipartisan push for security

Representatives Pat Harrigan of North Carolina and George Whitesides of California introduced the AI Cyber Defense Act on October 22. The bill seeks to establish a formal framework for protecting American artificial intelligence models from foreign exploitation. It also aims to leverage AI technology to bolster the cybersecurity of essential national infrastructure.

The legislation focuses on a dual-track approach to security. First, it seeks to harden the models themselves against theft. Second, it aims to empower the operators of critical systems to use AI as a shield against modern cyber threats. By involving the National Institute of Standards and Technology (NIST), the bill intends to create a standardized set of rules for the industry.

The bill was referred to the House Committee on Science, Space, and Technology. This committee will be responsible for reviewing the proposal before it can move further through the legislative process. The goal is to ensure that the United States stays ahead of adversaries who are looking to steal technological progress.

The AI Cybersecurity Roundtable

A central component of the bill is the creation of an AI Cybersecurity Roundtable. This group would be tasked with developing voluntary standards and best practices for the defensive use of AI. The roundtable is designed to bring together the primary players in the technology space to ensure the standards are practical and effective.

The bill specifies that the roundtable must include representation from both AI companies and owners of critical infrastructure. This includes the people responsible for managing power grids, hospitals, and financial systems. By bringing these groups together, the bill hopes to bridge the gap between those who build the technology and those who must rely on it for public safety.

NIST would have 180 days after the bill is enacted to begin engaging with private-sector stakeholders. These stakeholders are defined broadly to include any business that develops, trains, deploys, sells, or licenses AI systems. It also covers companies providing the underlying computing infrastructure, data, or technical services required for AI development.

Defending against distillation

One specific threat the bill addresses is the concept of AI distillation attacks. These are methods used by outside actors to extract valuable capabilities from advanced AI models without having to train the models from scratch. The bill directs NIST to work with the industry to establish a clear definition for these types of attacks.

Once a definition is established, NIST and industry partners will work together to create standards for detecting and preventing these specific exploits. The bill requires NIST to publish these standards and best practices on its official website within one year of their completion. This ensures that the information is accessible to the broader private sector.

The legislation emphasizes the need for proactive defense against foreign actors. It seeks to prevent adversaries from stealing the progress made by American developers. By creating a standard way to identify and block distillation attacks, the bill aims to protect the intellectual property of the companies leading the AI race.

Scope and exclusions

The bill is intentionally broad in its definition of eligible private-sector stakeholders. It covers a wide range of entities, from large tech giants to smaller firms that provide the necessary data or technical services for AI deployment. This wide net is intended to ensure that the resulting standards are comprehensive and applicable across the entire supply chain.

However, the bill includes specific exclusions for entities of concern. It explicitly bars businesses headquartered in China, Russia, Iran, North Korea, or Cuba from participating in the process. It also excludes any companies whose parent companies are headquartered in those specific countries. This is intended to prevent foreign adversaries from influencing the development of American security standards.

By focusing on domestic and allied stakeholders, the bill aims to create a secure environment for innovation. It seeks to balance the need for open collaboration with the necessity of protecting national security. The goal is to create a framework that supports American companies while shielding the public from external cyber threats.

Timeline for implementation

The bill sets out a clear sequence for how these standards will be developed and released. NIST must begin the engagement process with stakeholders within 180 days of the bill's enactment. This initial phase is meant to gather input from the people who are actually building and using the technology on a daily basis.

Within one year of establishing the AI Cybersecurity Roundtable, NIST is required to publish the voluntary standards and best practices for defensive AI use. While the bill does not set a separate deadline for the actual development of the standards, it provides a clear window for the initial publication of the roundtable's work.

The ultimate goal is to provide a clear roadmap for the industry to follow. As Rep. Whitesides noted, the bill aims to help protect American AI development while supporting operators facing evolving threats. By establishing these standards, the government hopes to help the private sector build the necessary safeguards for the technology's rapid advancement.