🔍 Read the full analysis: Does The Report Show China’s AI Models Learned From GPT-6 Astra? on ThorstenMeyerAI.com
Get business pricing on office and shipping supplies
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
The Information reported on October 28 that DeepSeek and Moonshot AI relied heavily on outputs from OpenAI’s GPT-6 Astra to train models released this month. Reuters said the three companies did not immediately respond; the public report provides no figures or details about how the outputs were obtained.
The Information reported October 28 that Chinese AI labs DeepSeek and Moonshot AI relied heavily on answers from OpenAI’s GPT-6 Astra to train models released this month. Reuters summarized the report, naming Moonshot’s Kimi K3.1 Pro and DeepSeek’s V4.1 Flash update; none of the three companies had immediately responded to Reuters, so the claim has not been confirmed publicly by them.
The Information’s report, by Jing Yang, Juro Osawa and Qianer Liu, said Astra had become a target for Chinese AI labs seeking to extract capabilities from the model. Its publicly accessible opening described strong demand for Astra after its release in early September and said Chinese AI labs were among a growing customer base. The rest of the article is paywalled, limiting what can be independently checked from its public text.
Reuters reported that the two labs’ latest models “relied heavily” on Astra outputs. The models identified were Kimi K3.1 Pro, released October 23, and DeepSeek-V4.1-Flash-1017, released October 17. The source material does not provide a count of prompts or outputs, a description of the alleged collection process, or an estimate of how much the outputs affected either model’s performance.
The labs have offered other accounts of their training. Moonshot’s launch post attributed Kimi K3.1 Pro’s gains to substantially larger post-training compute and large-scale reinforcement learning. DeepSeek’s update note, as described in the source material, did not discuss training data. Those statements do not independently settle whether Astra outputs were also used. Reuters said OpenAI, DeepSeek and Moonshot did not immediately respond to its requests for comment.
Infographic 01 · Distillation in plain words
How distillation works, and where it crosses a line.
A big model teaches a small one by example. The technique is ordinary; the dispute is about whose model is the teacher and how the answers were obtained.
Send the teacher prompts
A strong “teacher” model gets a huge number of carefully chosen tasks.
Keep answers and reasoning
The step-by-step working is worth far more than the final answer alone.
Teach a student to imitate
A smaller or cheaper “student” model learns to reproduce the teacher’s moves.
Normal practice
The U.S. advisory calls it “a legitimate and useful technique”.
- A lab distills its own large model
- Goal: smaller, faster, cheaper versions
- Open models used within their licence
Where the fight starts
What OpenAI, Anthropic and U.S. agencies allege.
- The teacher is a rival’s model
- Industrial scale: millions of exchanges
- Fake, resold or stolen accounts and keys
- Breaks terms that ban building competing models
Why hidden reasoning is the prize
Anthropic’s own test: a model trained on Claude Opus 4.8 transcripts, scored on SWE-bench Verified (coding).
That is why providers now hide or summarize reasoning, and why extraction attempts target it.
Distillation is a student copying a top tutor’s worked solutions. Legal in a classroom. Contested when the tutor never agreed.Thorsten Meyer · my reading
Sources: NSA/CISA/FBI advisory AA26-251A, 8 Sep 2026; Anthropic, 12 Oct 2026 (vendor’s own experiment). Analogy and framing: my reading.
Powered by Thorsten Meyer AI
Why Model Training Provenance Matters
The report matters because using a rival’s outputs at scale could give a model developer access to capabilities built through another company’s research and computing investment. It also puts attention on training-data provenance: companies that adopt AI systems may want to know not only how a model performs, but what data and processes shaped it.
The underlying technique, called distillation, is not inherently improper. A developer can use answers from a larger “teacher” model to train a smaller “student” model, and labs also distill their own systems. The dispute described in the report is about whether a competitor’s model was used, how extensively, and whether access was obtained in ways that breached contractual terms or involved deception. The public evidence supplied so far does not answer those questions.
There may also be a commercial consequence if the allegations are substantiated. The source material lists Astra at $10 per million input tokens and $50 per million output tokens, compared with $2 and $10 for Kimi K3.1 Pro and $0.30 and $1.20 for DeepSeek’s model at peak hours. It also gives respective scores of 57, 46 and 32 on an unnamed independent index. These figures suggest lower prices alongside lower scores on that index; they do not establish that distillation caused the price or performance differences.
For customers, the immediate practical issue is due diligence rather than a confirmed change to product access. Businesses can check vendors’ disclosures about data sources and model development, test systems on their own work, and maintain alternatives. The source material also advises users to avoid cheap, unofficial API resellers, though it does not report that any particular reseller was involved in this case.
Popular right now
De'Longhi Rivelia Fully Automatic Espresso Machine with Milk Frother, White
Breville BES870XL Barista Express Espresso Machine, Brushed Stainless Steel
Samsung 42-Inch S90H OLED 4K Smart TV Bundle
As an affiliate, we earn on qualifying purchases.
Earlier Allegations Set the Backdrop
The current report follows earlier public allegations about efforts to obtain model outputs. The source material says that in February 2026, Anthropic alleged that DeepSeek, Moonshot and MiniMax generated more than 16 million exchanges with Claude through accounts associated with the companies. It also says that in September, OpenAI linked a “core cluster” of a July campaign to extract hidden reasoning from its models to individuals associated with Moonshot AI.
According to the source material, U.S. security agencies named DeepSeek and Moonshot in an advisory, while China’s commerce ministry called the accusations baseless. These earlier statements provide context but do not prove the separate claim about Astra training. The source material also reports that a U.S. government advisory described distillation as a legitimate and useful AI research technique, underscoring the distinction between the method itself and alleged ways of using or accessing a competitor’s service.
A separate account from South Korean broadcaster SBS said sources told it Chinese labs had previously focused on Anthropic’s Claude and shifted toward OpenAI after Anthropic tightened defenses. SBS also quoted an OpenAI spokesperson saying the company takes the issue seriously and continues working on it. The original paywalled report’s details could not be checked against those additional claims, so they should be treated as attributed reporting, not independently established facts.
“DeepSeek and Moonshot “relied heavily on outputs” from GPT-6 Astra to train their latest models.”
— The Information, as summarized by Reuters
Key Details Remain Unverified
The publicly available account does not say how many Astra outputs were used, which prompts were involved, or whether the labs accessed the model directly, through third parties, or by another route. It also does not establish the role of hidden reasoning transcripts, which are distinct from ordinary model answers and are not described in the available account of this allegation.
There has been no public confirmation or denial from OpenAI, DeepSeek or Moonshot in the Reuters account cited here. It is also unclear whether OpenAI has opened a formal investigation, whether it has identified specific accounts or transactions, or whether any regulator or court is examining this particular report. The allegation should not be treated as a finding of contractual or legal wrongdoing.
Responses and Evidence to Watch
The next developments to watch are direct responses from the three companies and any further reporting that discloses evidence behind the claim. Relevant details would include the scale and timing of the alleged use, how access was obtained, and whether the companies dispute the characterization of their training process.
OpenAI’s terms, as described in the source material, prohibit using its outputs to build competing models. Whether those terms apply to the conduct alleged, and whether they were breached, would depend on facts not available in the public account. Until more information emerges, users and businesses should distinguish the reported allegation from confirmed findings. No immediate change to the named models or their availability is reported in the source material.
Key Questions
What does the report allege?
The Information reported that DeepSeek and Moonshot AI relied heavily on GPT-6 Astra outputs to train recent models. The companies had not publicly confirmed the claim in Reuters’ account.
Which models are named?
The report names Moonshot’s Kimi K3.1 Pro, released October 23, and DeepSeek’s V4.1 Flash update, released October 17.
Is distillation itself improper?
No. Distillation is a common training technique, including for creating smaller versions of a lab’s own model. The reported dispute concerns alleged use of a competitor’s outputs and whether access or use violated applicable terms.
Has either AI lab confirmed using Astra outputs?
Not in the accounts provided here. Reuters reported that OpenAI, DeepSeek and Moonshot did not immediately respond to its requests for comment.
Does the report prove a legal or contract violation?
No. The available reporting presents an allegation, not a legal finding. The public material does not provide enough detail about how the outputs were obtained or used to determine whether any agreement or law was violated.
Source: ThorstenMeyerAI.com
