OpenAI Accuses Chinese Firm of Stealing AI Model Data
Left 50%
Center 0%
Right 50%
1 left · 0 center · 1 right
What happened
OpenAI said it identified and disrupted a coordinated campaign, beginning in early July, in which operators tried to extract protected reasoning from its artificial intelligence models by manipulating interactions rather than by breaching encryption, databases, or stored user conversations. The activity later surged to about 16,000 requests from more than 4,000 users over two days and involved related activity across a cluster of more than 15,000 users; OpenAI said it shut down or limited the accounts, tightened sign-ups, and had fully disrupted the campaign by July 28. The company described the method as “adversarial distillation,” in which one model’s outputs or reasoning are used to train or improve another, and attributed a core cluster of the activity to individuals associated with Chinese startup Moonshot AI, the developer of the Kimi chatbot, while saying it was unclear whether all operators originated from a single actor. OpenAI said it shared its findings with other developers through the Frontier Model Forum and with government information-sharing channels; Moonshot AI did not immediately respond to requests for comment.
Sourcing check · Fully on the record · 2 articles
2 of 2 articles cite a named source for the story. It was on the record from the first report.
What these terms mean and how they're counted
- On the record · names a source 2 of 2
- On the record means the article cites a source by name for the story's main claim: a person, or an institution speaking officially, such as a company statement or a named agency's spokesperson. The opposite is an unnamed source. The count is how many of the counted articles cite at least one such source; the names are everyone named across them, each once however the articles spell it. Here: OpenAI and Thomas Duesterberg
- Primary evidence · cites a document or data 1 of 2
- Primary evidence means the article cites a record you could check yourself: a court filing or ruling, a dataset or report, a recording or transcript, an official statement or press release, market data. A quote is not a document. The count is how many articles cite one. Here: company report, company statement
- Credits another outlet 0 of 2
- The article says the story comes from another outlet's reporting (“first reported by Reuters”). Those articles relay the story; they are not independent confirmation. Here: Each article presents its own reporting.
- Unnamed sources 0 of 2
- The main claim is attributed to people who are not identified: “people familiar with the matter”, “officials said”. The count is how many articles do this.
- How it developed
- When the first report landed and whether it named anyone; when the first named source and the first document appeared. Here: First report Oct 1, 02:14 UTC by CNBC, naming its source. First document 19 hours later, from Daily Caller (company report).
- Not counted 0
- Opinion, analysis and video pieces, and articles too short to assess. Every count above is out of the counted articles. An article republished under a new headline is one article.
This is about who is cited, not whether the story is true. Each source below says what it names. How we count
Omitted — what each side leaves out
Unpacked
The operators never broke into OpenAI, a limit right-leaning coverage leaves out. Left-leaning coverage at CNBC reports they did not breach encryption, databases, or stored conversations and instead manipulated ordinary model chats to reproduce hidden reasoning. Without that fact, the Daily Caller's "red-handed" framing reads as a break-in rather than a campaign of manipulated queries.
Right-leaning coverage leads with a Chinese firm trying to "copy its homework." Left-leaning coverage leads with the AI race heating up and never uses theft language at all. The first words a reader sees already pick a verdict.
Did the queries actually reproduce OpenAI's hidden reasoning, and is query-based distillation even illegal?
Bottom line
Only CNBC reports that OpenAI's stored conversations were never touched, so a Daily Caller reader is left with a break-in story the company itself did not tell.
The receipts — 1 fact one side's coverage lacks
- OpenAI said the operators did not breach encryption, databases, or stored user conversations
The Left View
CNBC frames the report as a sign the “AI race heats up” and as feeding “growing concerns among U.S. AI companies that rivals could use their models to develop competing technology more quickly and cheaply.” It places the episode weeks after Anthropic accused Chinese developers including Moonshot AI and Alibaba of secretly using Claude to help train their own systems, treating both as industry-copying pressure rather than a settled theft case. The account stays with OpenAI’s wording of an “alleged model-copying campaign” and a linked “core cluster,” and centers OpenAI’s warning that extracted reasoning “could allow others to reproduce advanced capabilities without making the same investment in developing and safeguarding frontier models,” posing “potential safety and national security risks.”
The Right View
The Daily Caller frames the same report as OpenAI catching a Chinese firm “red-handed” in a “15,000 phony account scheme” to “copy its homework,” and treats the terms-of-use violation as theft. Hudson Institute fellow Thomas Duesterberg said it “indicates that the Chinese firms, which claim to be almost as powerful as the American firms, really can’t compete at all unless they steal our stuff,” and called OpenAI’s language “a polite way of saying they’re stealing the stuff.” He argued Chinese models are built to repeat government-approved narratives, will not answer questions about topics such as Tiananmen Square, and could leave American users “led astray, or totally misled,” adding that the operators are “disciplined and they’re relentless.” The piece also cites a similar OpenAI inquiry into DeepSeek in January 2025 and records Duesterberg’s call to ban those models in the United States and ban U.S. capital investment in Chinese firms.
Our Take (balanced)
The left-leaning account’s strongest argument is that OpenAI itself described distillation via manipulated queries, not a systems intrusion, and would not say all operators were one actor—so the best-supported reading is competitive copying in a two-firm-ecosystem race, reinforced by the parallel Anthropic accusations rather than by proof of a unified heist. The right-leaning account’s strongest argument is OpenAI’s attribution of a core cluster to Moonshot-associated individuals targeting “hidden reasoning,” which Duesterberg treats as going after the core of frontier models and as evidence Chinese labs “really can’t compete at all unless they steal our stuff.” The central unresolved tension is whether “adversarial distillation” at this scale is aggressive commercial model-copying inside a global AI race, or intellectual-property theft that exposes a structural gap between Chinese labs and U.S. frontier models.
2 sources
More stories
The week's bottom lines, in your inbox
One email a week: the five stories that mattered and what they actually mean. Free.