A bipartisan group of U.S. lawmakers has introduced the Countering Adversarial Threats and Security Risks Act, or CATS Act, in a push to stop foreign rivals from obtaining leading American artificial intelligence technology at low cost. The proposal was announced in a news release from Senator Adam Schiff’s office, which said Schiff, Senator Jim Banks, Representative Bob Latta and Representative George Whitesides jointly put forward the bill.
The measure is designed to let U.S. AI labs and security personnel share threat intelligence without running afoul of antitrust law, with a focus on major national security threats tied to foreign actors, including so-called AI distillation attacks.
The bill is aimed at AI distillation attacks
The release describes an AI distillation attack as a process in which foreign laboratories systematically send millions of queries to the most advanced U.S. AI models, extract large amounts of data, and then use that information to train their own advanced models at far lower cost.
Supporters of the bill argue that this amounts to low-cost reverse engineering of U.S. technology. In their view, it threatens America’s lead in building the most powerful and efficient AI systems and undercuts the advantage created by large investments in the sector.
On July 23, Taipei time, Schiff and Banks joined Latta and Whitesides in formally introducing the CATS Act. The bill’s central goal is to strengthen the ability of U.S. AI labs to defend against foreign adversaries carrying out AI distillation attacks.
“We cannot allow China or any foreign adversary to steal America’s advances in AI technology and then use those technologies to undermine our national security,” Schiff said. Banks said the AI race is one the United States cannot afford to lose, adding that the Chinese Communist Party tries every day to steal technology and that the bill would give U.S. AI companies a lawful path to cooperate in defending against those efforts.
Statutory exemption would allow threat sharing
A central feature of the bill is a statutory exemption meant to clear the way for coordination in specific circumstances. The report says AI developers that coordinate privately today could face antitrust scrutiny. Under the CATS Act, non-federal entities would be allowed to share information in good faith for the purpose of preventing or mitigating covered AI security risks.
The bill also says that, if the sole purpose is to reduce risk and prior written notice is provided, companies could coordinate to delay or limit the deployment of high-risk AI models.
The lawful coordination channels listed in the proposal include the National Security Agency’s AI Security Center, the National Institute of Standards and Technology’s CAISI, and industry groups such as the Frontier Model Forum.
Modeled in part on cybersecurity information sharing
According to the report, the legislative approach draws on the Cybersecurity Information Sharing Act, which it describes as having operated successfully for 10 years. Backers want to extend the government-industry information-sharing model used in cyber threat defense into AI security.
Safeguards and endorsements
The bill also includes guardrails intended to prevent the exemption from being used for anti-competitive conduct. It would allow the attorney general to seek injunctive relief against entities that misuse the exemption.
The proposal has already received support from several companies and organizations named in the report, including Google, Andreessen Horowitz, the Center for AI Safety Action Fund, and the Software & Information Industry Association. The article describes the bill as an attempt to balance national security defenses with continued industry innovation.