'This isn't competition, it's theft’: Cohen urges DOJ to expose China’s ‘stealth’ strategy in Senate

Key Concepts

• IP Theft (Intellectual Property Theft): The unauthorized use or appropriation of protected creative works, trade secrets, and proprietary data.
• AI Distillation: A process where actors query advanced "frontier" AI models to extract knowledge or patterns, effectively "stealing" the capabilities of the original model to train their own.
• NIL (Name, Image, and Likeness): The rights of entertainers and public figures to control the commercial use of their identity, currently threatened by AI-generated deepfakes.
• Frontier Models: The most advanced, large-scale AI systems currently in development.
• Defend Forward: A proactive cybersecurity strategy involving the identification and disruption of threats before they reach domestic networks.
• Antitrust Guidance: Legal frameworks that currently discourage AI companies from sharing threat intelligence due to fears of collusion or anti-competitive behavior.

---

1. The Nature and Scope of IP Theft

The discussion highlights a systemic issue where Chinese entities engage in the "scooping and scraping" of American intellectual property. This practice is described as being historically tolerated or encouraged by the Chinese government to accelerate economic development.

• AI-Driven Theft: Concerns are centered on the use of stolen IP to train Large Language Models (LLMs). Examples include the unauthorized use of content to create deepfakes (e.g., AI-generated brawls involving Tom Cruise and Brad Pitt) and the manipulation of copyrighted media (e.g., Stranger Things).
• Strategic Objectives: Beyond commercial gain, the CCP is accused of using stolen data to rip off trade secrets, acquire military information, and facilitate population indoctrination.

2. Regulatory and Legal Challenges

• The "Double Disclosure" Dilemma: Companies are often reluctant to report trade secret theft to the Department of Justice (DOJ) or other agencies because the legal process requires disclosing sensitive information to the court, which may lead to further exposure of the stolen data to the original perpetrator.
• Chinese Legal Framework: Interestingly, the witness notes that Chinese law actually prohibits the use of content without a legal source for AI training. There is a suggestion that the U.S. could leverage China’s own regulatory bodies—specifically the Copyright Bureau and the Ministry of Propaganda—to enforce these existing, albeit ignored, laws.
• Legislative Action: The "Countering Chinese Espionage Reporting Act" (introduced by Senators Blackburn and Coons) aims to mandate an annual DOJ report on Chinese espionage threats to improve transparency and tracking.

3. AI Distillation and Technical Defense

Senator Schiff raised concerns regarding "distillation," where Chinese actors use high-volume, automated queries to extract the intelligence of U.S. frontier models.

• The Company Dilemma: AI companies face a trade-off between security and usability. Over-restricting access to prevent distillation can drive away legitimate users and reduce profitability.
• Government Role:
  • Antitrust Reform: The witness argues that the government should clarify antitrust guidance to allow AI companies to share threat intelligence and coordinate defense strategies without fear of legal repercussions.
  • Centers of Excellence: The AI Security Center (NSA) and the Center for AI Standards and Innovation (NIST) are identified as critical hubs for sharing classified threat intelligence with private companies.
  • Intelligence Toolkit: The witness suggests utilizing the full suite of FBI and Intelligence Community (IC) tools to determine if these campaigns are coordinated by the state or carried out by independent actors.

4. Notable Statements

• Senator Blackburn: "They’re very concerned about what China is doing to scoop and scrape their intellectual property and train LLMs on it and then poof it is gone."
• Witness (on Chinese Law): "A lot of Chinese companies do this opportunistically and they're actually violating Chinese law which requires a legal source for content that is used to train AI products."
• Witness (on Antitrust): "The companies could coordinate with each other, could share information about both what they're detecting and how they're preventing distillation. Right now they're holding back on that because they're concerned about antitrust."

5. Synthesis and Conclusion

The testimony establishes that IP theft has evolved from traditional industrial espionage to a sophisticated digital threat involving AI distillation and the exploitation of NIL rights. The primary takeaways are:

1. Transparency is lacking: Due to the fear of "double disclosure," the true scale of IP theft remains hidden.
2. Regulatory Leverage: There is potential to use China’s own internal copyright laws as a diplomatic and legal lever.
3. Need for Collaboration: The government must bridge the gap between classified intelligence and private sector defense by providing legal safe harbors (antitrust relief) for companies to share data on AI misuse.
4. Proactive Defense: Moving toward a "defend forward" posture is essential to identifying whether these AI-based attacks are state-sponsored or decentralized.