OpenAI accuses DeepSeek of stealing tech for chatbot

In a significant escalation of the AI arms race between the U.S. and China, OpenAI has officially accused the Chinese startup DeepSeek of utilizing “sophisticated and obfuscated methods” to steal intellectual property. In a memo sent to the House Select Committee on China on February 12, 2026, OpenAI alleged that DeepSeek “distilled” its frontier models to bypass costly research and development.

The controversy centers on how DeepSeek managed to produce its R1 model—which rivals GPT-4o in reasoning—for a reported training cost of just $5.6 million, a fraction of the hundreds of millions spent by U.S. labs.

The Core Allegation: “Illegal Distillation”

OpenAI’s memo describes a process called distillation, where a smaller AI model (the “student”) is trained using the outputs of a larger, more advanced model (the “teacher”). While distillation is a common technique in AI research, OpenAI asserts that DeepSeek’s execution was unauthorized and violated its terms of service.

Key points from the OpenAI memo:

• Bypassing Safeguards: OpenAI claims DeepSeek used “new, obfuscated methods” to hide its activity and scrape data from OpenAI’s API without detection.
• “Free-Riding” on R&D: The memo argues that DeepSeek is “free-riding” on billions of dollars of American infrastructure and research investment.
• National Security Risks: OpenAI warned lawmakers that when models are distilled, the built-in safety filters and alignment protocols of the original model are often stripped away, potentially allowing the “copycat” model to be misused for high-risk tasks like biology or cyber-warfare.

The Industry Reaction: A “Peak of Hypocrisy”?

The accusations have sparked a firestorm within the tech community. While some lawmakers have called for immediate investigations into DeepSeek’s practices, others have pointed out a glaring irony.

Critics and legal experts have noted that OpenAI itself has faced dozens of lawsuits from authors, artists, and news organizations (including The New York Times) for allegedly scraping the entire internet to train its models without permission.

“It’s roughly the equivalent of a school bully complaining to the teacher that his stolen lunch was stolen from him by another bully,” noted one industry commentator.

What’s Next for DeepSeek?

Despite the allegations, DeepSeek remains a formidable competitor. The startup has consistently denied any wrongdoing, maintaining that its efficiency comes from architectural innovations like Multi-head Latent Attention (MLA) rather than theft.

DeepSeek is expected to release its next flagship model later this month, which will likely further test the tension between open-source innovation and corporate intellectual property rights.

Comparison of Model Efficiency