Leveraging LLMs to Outperform Competitors • The Register

Taylor Keatsman

Published: February 14, 2026 5:52 PM ET

Leveraging LLMs to Outperform Competitors • The Register

Two prominent AI companies, Google and OpenAI, have raised concerns about potential threats to their intellectual property. This week, they warned that individuals and organizations, including the Chinese firm DeepSeek, are attempting to replicate their large language models (LLMs). Such actions can lead to significant intellectual property theft, as explained by John Hultquist, a chief analyst at Google’s Threat Intelligence Group.

Threats from Competitors

Hultquist indicated that these threats come from various actors worldwide. He pointed out that private-sector companies are primarily behind these efforts, although he refrained from naming specific organizations or nations. According to him, the underlying logic of LLMs represents valuable intellectual property. If competitors successfully distill this logic, they can replicate complex AI technologies.

Distillation Attacks

Google has described the phenomenon of replicating its models through specific prompts as “distillation attacks.” A recent report highlighted a campaign that involved over 100,000 prompts aimed at mimicking the reasoning capabilities of Gemini, Google’s LLM. This cloning process is not just costly but also threatens the competitive landscape for American tech firms, who have invested billions in developing these technologies.

Distillation attacks exploit legitimate access to AI models.
Competitors can create AI systems with reduced development expenses.
Google has implemented measures to detect and safeguard its models from these threats.

Legal and Enforcing Measures

Google has established legal frameworks to combat unauthorized distillation. Violating its terms of service can lead to account suspension or legal action against perpetrators. Nonetheless, the very nature of LLMs allows for certain vulnerabilities. Public access to these models complicates the enforcement of protective measures.

The Growing Risk

Hultquist warned that as more companies share their models, the risk of distillation attacks will escalate. Even organizations outside the tech sector, such as financial institutions, may fall victim to these tactics. OpenAI has also voiced its concerns regarding distillation, specifically highlighting DeepSeek and other Chinese entities for their role in copying American AI innovations.

Challenges and Responses

OpenAI’s analysis emphasizes that China’s techniques for distillation have evolved, incorporating multi-stage operations and sophisticated methods. In a memorandum to the House Select Committee on China, OpenAI acknowledged that it has begun enhancing its detection mechanisms. The company is proactive in enforcing its terms of service by banning accounts engaged in unauthorized activities.

Stronger detection systems are in development to combat distillation.
Proactive measures include removing users attempting to distill models.

Despite these efforts, OpenAI recognizes that a collective approach is essential for effective protection against distillation threats. The company has called for collaboration with the US government to establish best practices and improve overall security within the industry.

Call for Legislative Support

In its memo, OpenAI urged Congress to close API router loopholes. This would prevent entities like DeepSeek from gaining unauthorized access to US technologies and cloud resources. As AI technologies continue to evolve, safeguarding these assets is crucial for maintaining a competitive advantage in the global market.