🔍 Read the full analysis: How To Read The Report On China’s AI Models And GPT-6 Astra on ThorstenMeyerAI.com
Get hardware and tech essentials delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
The Information reported on Oct. 28 that DeepSeek and Moonshot AI relied heavily on OpenAI’s GPT-6 Astra outputs to train recent models. Reuters repeated the report, but the companies did not respond to its request for comment, and public evidence about the alleged data, methods and scale remains limited.
The Information reported on Oct. 28 that Chinese AI companies DeepSeek and Moonshot AI relied heavily on outputs from OpenAI’s GPT-6 Astra to train models released this month. Reuters summarized the report within hours, but said OpenAI, DeepSeek and Moonshot did not immediately respond to requests for comment, and the public record does not show how the outputs were obtained or how much they contributed.
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 its capabilities. Its public opening describes Astra as OpenAI’s most powerful released model and says demand has grown since its early-September release; the rest of the article is paywalled. Reuters reported the central allegation and named Moonshot’s Kimi K3.1 Pro and DeepSeek-V4.1-Flash-1017 as models that benefited from distillation. Kimi K3.1 Pro was released Oct. 23, while DeepSeek’s update launched Oct. 17.
The allegation is not supported by public technical evidence in the available reporting. The source account does not provide prompt counts, training records, access methods or a measured contribution from Astra outputs. Moonshot’s launch post credited Kimi’s gains to substantially greater post-training compute and large-scale reinforcement learning. DeepSeek’s update note did not describe its training data. Those statements do not independently confirm or disprove the reported use of Astra.
Reuters said it had not received an immediate response from the three companies. ThorstenMeyerAI.com also cited Korean broadcaster SBS as reporting that labs had shifted from Claude to Astra after Anthropic tightened defenses, and as quoting an OpenAI spokesperson saying the company takes the issue seriously. The source author said those details could not be checked against The Information’s original report. They should be treated as secondary reporting, not independently verified evidence.
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
If the report is accurate, it would put attention on how a model provider’s outputs may be used to build competing systems. The concern is not that one model learns from another in itself: distillation is a standard training technique. The disputed issues are whether the teacher model belongs to a rival, whether outputs were collected at industrial scale, and whether access violated provider rules or involved deceptive accounts.
The commercial stakes include the value of advanced model capabilities and the cost of reproducing them. ThorstenMeyerAI.com 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. The article also cites an independent index scoring the three at 57, 46 and 32, respectively. Those figures indicate different prices and scores, not equivalent performance; the source does not specify enough about the index here to draw broader conclusions.
For businesses evaluating AI vendors, the report is a reminder to ask about training-data provenance, terms compliance and model fallback plans, then test systems on their own workloads. For users, the source says no immediate app changes were known. It advises caution around cheap, unofficial API resellers, whose access and data handling may be unclear.
Products Worth Considering
Apple MacBook Pro 16.2-inch with M5 Pro, 18-core CPU, 20-core GPU, 48GB Memory, 1TB SSD, Nano-Texture Display, Space Black
Apple MacBook Pro 14.2-Inch Laptop with M2 Pro, 16GB Memory, 1TB SSD, 19-Core GPU, Space Gray
Apple MacBook Pro 14-inch (2021) – M1 Pro Chip, 10-Core CPU, 16-Core GPU, 16GB RAM, 1TB SSD, Space Gray
As an affiliate, we earn on qualifying purchases.
Distillation and Earlier Allegations
In distillation, a larger “teacher” model produces answers to prompts, and a smaller “student” model is trained to imitate them. The technique can make a model cheaper to run and is used by AI labs to create smaller versions of their own systems. A U.S. government advisory cited in the source material describes distillation as “a legitimate and useful technique in AI research.” That general legitimacy does not settle whether a particular use complies with a provider’s contract or applicable law.
The report arrives amid earlier disputes over model access. ThorstenMeyerAI.com says OpenAI stated in September that it linked a “core cluster” in a July campaign to extract hidden reasoning from its models to individuals associated with Moonshot AI. It also says U.S. security agencies named both DeepSeek and Moonshot in an advisory, while China’s commerce ministry called the accusations baseless. These are prior statements and allegations; they do not independently establish the new claim about Astra training data.
OpenAI’s terms, as described in the source, prohibit using its outputs to build competing models. Whether the reported conduct occurred, who obtained any outputs and whether the terms were breached are unresolved. Any legal conclusion would depend on facts and applicable law not established in the public reporting.
““a legitimate and useful technique in AI research.””
— U.S. government advisory, as quoted by ThorstenMeyerAI.com
What the Public Record Does Not Show
The available reporting does not establish how many Astra outputs were used, how they were collected, or whether any accounts were falsified or resold. It also does not provide training logs, independent model analysis or a company confirmation. The Information’s detailed account is behind a paywall, while Reuters’ public summary gives the allegation but not the underlying evidence.
It is also unclear how much any alleged Astra data contributed to either model’s capabilities compared with other training methods. Moonshot has cited additional compute and reinforcement learning for Kimi’s gains; DeepSeek’s update note does not disclose its data. The companies’ lack of an immediate response is not confirmation of the allegation. No public legal finding about this reported conduct is cited in the source material.
Responses and Evidence to Watch
The next useful developments would be direct responses from OpenAI, DeepSeek and Moonshot, along with any technical evidence that clarifies the source, volume and use of model outputs. Company statements, disclosures about training practices, or further reporting based on verifiable records could sharpen what is currently a serious but incompletely documented allegation.
Until then, readers and organizations should distinguish the reported claim from established facts. The immediate practical question for model users is not whether to infer misconduct from the article, but whether a provider can explain its data practices, account access and terms. Any wider consequences—including contract disputes, policy action or changes to model access—remain dependent on evidence and on what the companies or relevant authorities do next.
Key Questions
What did The Information report?
It reported that DeepSeek and Moonshot AI relied heavily on GPT-6 Astra outputs to train recent models, including DeepSeek-V4.1-Flash-1017 and Moonshot’s Kimi K3.1 Pro. The companies have not confirmed the claim in the cited coverage.
Does the report prove that the companies violated OpenAI’s terms?
No. The public reporting cited here does not establish how the outputs were obtained or whether a particular contract was breached. The source says OpenAI’s terms prohibit using its outputs to build competing models, but no legal finding about this case is cited.
Is distillation itself improper?
Not inherently. Distillation is a common method in which a smaller model learns from a larger model’s outputs. The dispute concerns the alleged use of a rival’s model and whether the collection or use of its outputs violated terms or involved deceptive access.
Have OpenAI, DeepSeek or Moonshot confirmed the allegation?
Not in the cited reporting. Reuters said the three companies did not immediately respond to its requests for comment. That lack of response does not establish whether the allegation is true.
Source: ThorstenMeyerAI.com
