Live Nov 18 · Cohort 3Nov 18, 19:00 MSK — free hands-on AI agents live workshop · Live Cohort 3 starts Nov 23Join Nov 18 Live →Cohort 3 (Nov 23)
ReleaseInsight · On RadarBloomberg Tech→ original

OpenAI accuses DeepSeek of using data from American models to train R1

OpenAI has warned US authorities about unfair competition from the Chinese startup DeepSeek. According to an internal memo, DeepSeek uses sophisticated distillation methods to extract data from advanced American neural networks. These outputs are allegedly used to train the next generation of the R1 chatbot. OpenAI claims this approach gives the Chinese side an unfair advantage. The incident highlights the escalating technology race between the US and China, and raises questions about protecting intellectual property in the development of large language models.

AI-processed from Bloomberg Tech; edited by Hamidun News
OpenAI accuses DeepSeek of using data from American models to train R1
Source: Bloomberg Tech. Collage: Hamidun News.
Listen to article

# OpenAI accuses DeepSeek of unauthorized use of American models

Tension in the global AI race has reached a new level. OpenAI has sent American lawmakers an official warning that the Chinese startup DeepSeek is using sophisticated, cunning methods to extract useful information from leading American neural networks in order to train its breakthrough chatbot R1. The incident, reported by Bloomberg, reveals more than a commercial dispute: it is a fundamental clash over intellectual property protection and competitive advantage in the age of artificial intelligence.

According to an internal OpenAI memo, DeepSeek uses distillation, a process in which a smaller model learns from the outputs of a larger, more advanced system. In practice, this means DeepSeek engineers systematically send queries to American models, analyze the responses and use the resulting data to train their own algorithms. This approach lets the Chinese company save enormous amounts of compute and development time that are normally required to build models with the same level of performance. In other words, by OpenAI's account, DeepSeek is essentially free-riding on American companies' investments in research and development.

The context of this conflict is critical for understanding its scale. Developing cutting-edge language models takes hundreds of millions of dollars for computing hardware, hiring top specialists and running experiments. OpenAI, Microsoft and other American giants have spent billions to reach their current level of performance. DeepSeek, by contrast, managed to reach a comparable level in far less time and at far lower cost. That raises legitimate questions about whether this success was achieved fairly or is the result of systematically "borrowing" knowledge from American models.

OpenAI's accusation also matters because it puts the spotlight on the vulnerability of API interfaces. Most leading models are available through web services where developers can send queries and receive responses. Although the terms of use prohibit using outputs to train competing systems, this is extremely hard to detect and prevent. DeepSeek could potentially have simply run this scheme at scale, processing millions of queries and extracting structured information from them for its own needs.

The incident also highlights a growing imbalance in the technology race between the US and China. American companies face strict restrictions on exporting advanced technology to China, but they cannot fully cut off access to their public services without harming their global business. DeepSeek, it seems, has found a gap in this system. OpenAI is now calling for stronger intellectual property protection and tighter monitoring of how commercial APIs are used.

What happens next depends on how American regulators react and whether they take steps to protect the interests of the domestic industry. In the short term, this may lead to tighter access conditions for advanced models and more sophisticated verification systems. In the long term, the conflict underlines the need to rethink how intellectual property is divided in an era of open APIs and a global digital economy. How to protect innovation and investment while staying open to scientific progress remains one of the industry's sharpest questions.

Story
ZK
Zhemal Khamidun — CPO at Alpina Digital, product manager of AlpinaGPT. Building AI products since 2019. Hamidun News is a pet project that grew into a production aggregator with 6,000+ readers.
✦ Unified AI Workspace · Nomad

Test Frontier LLMs & Autonomous Agents in One Workspace

Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.

Try Nomad AI →Launch Nomad Agent
What do you think?
Loading comments…