OpenAI accuses DeepSeek of copying US chatbot model

OpenAI has informed US lawmakers that Chinese AI company DeepSeek is using advanced distillation techniques to train its R1 chatbot, according to a memo reviewed by Bloomberg.

The memo alleges DeepSeek copied capabilities from US models despite OpenAI’s efforts to prevent this.

OpenAI raised concerns shortly after the R1’s launch last year, prompting an internal investigation with Microsoft.

The company says DeepSeek used obfuscated methods to evade protections and accessed US models through third-party routers and unauthorized resellers.

OpenAI also warned that copying models could weaken safeguards and increase misuse risks on sensitive topics.

DeepSeek did not immediately respond to requests for comment.

The US government has previously scrutinized DeepSeek’s chip sourcing and export compliance amid broader concerns over China’s AI development and hardware access.

🔗 Source: Bloomberg

🧠 Food for thought

Implications, context, and why it matters.

DeepSeek’s performance and pricing rattled markets

DeepSeek released its R1 model, then Nvidia’s market capitalization dropped by nearly $600 billion. Investors worried that more efficient models could cut demand for high-end chips 1.
Benchmark results put R1 on par with, or ahead of, OpenAI’s o1 on reasoning tests such as MATH-500 and AIME 2024 1.
DeepSeek also undercuts on cost. The R1 API lists $0.55 per million input tokens and $2.19 per million output tokens, while OpenAI’s o1 pricing is cited at $15 per million input tokens and $60 per million output tokens 2.

US policy may be speeding Chinese innovation while a legal gray area hangs over AI

Distillation, which trains a smaller model to copy a larger model’s behavior, sits in a legal gray area. The dispute could become a test for AI similar to Oracle v. Google for software APIs, which are the rules for how software components talk to each other 3.
A court win for distillation could open AI development to smaller firms that want specialized models. A loss could lock in the edge held by big players such as OpenAI 3.
US export controls meant to slow China’s AI progress may instead push efficiency gains by limiting access to top chips. That shift could turn into a long-term strategic advantage for Chinese firms 4.
Enforcing US terms of service in China remains hard, so conflicts may move toward national security policy. OpenAI alleges DeepSeek used obfuscated methods to evade protections, then reached US models via third-party routers and unauthorized resellers 5.