🔍 Read the full analysis: Could China’s Latest AI Models Have Learned From GPT-6 Astra? on ThorstenMeyerAI.com
Get business pricing on monitors, keyboards and dev gear
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
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. The claim is based on unnamed sources, and the three companies had not confirmed or denied it to Reuters. The method described is model distillation, which is common in AI development; the disputed issues include use of a competitor’s outputs, scale and how access was obtained.
The Information reported on Oct. 28 that China-based AI labs DeepSeek and Moonshot AI relied heavily on outputs from OpenAI’s GPT-6 Astra to train recent models, a claim that could raise questions about model-training practices and OpenAI’s terms. Reuters relayed the report, but said OpenAI, DeepSeek and Moonshot did not immediately respond to requests for comment.
The models named in the report are Moonshot’s Kimi K3.1 Pro, released Oct. 23, and DeepSeek-V4.1-Flash-1017, released Oct. 17. The Information cited people with direct knowledge, according to its report and Reuters’ account. The Information’s public-facing summary said Chinese AI labs were among a growing group of customers seeking to extract capabilities from Astra, which it described as OpenAI’s most powerful released model. The article’s full text is behind a paywall.
Public evidence about the alleged training is limited. The reports do not provide figures for how many Astra outputs were used, the prompts or process involved, or how the labs obtained access. The launch materials cited in the source article offer different explanations for model development: Moonshot attributed Kimi K3.1 Pro’s gains to substantially larger post-training compute and large-scale reinforcement learning, while DeepSeek’s update note did not describe its training data. Those statements do not independently settle whether Astra outputs were used.
A separate account from South Korea’s SBS, published Oct. 29, said sources described a shift from using Anthropic’s Claude to OpenAI models after Anthropic tightened defenses. SBS also reported that an OpenAI spokesperson said the company takes the issue seriously and continues working on it. The source article says those details could not be checked against The Information’s full report. They should be treated as secondary reporting, not independently confirmed findings.
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
The Dispute Over Training Data
If the report is accurate, it would place a familiar AI training method at the center of a dispute about whose model outputs may be used and under what conditions. Distillation can help produce models that are less costly to run, potentially making capable AI tools available at lower prices. But using a rival’s paid outputs at large scale may raise contractual and access questions, especially if the provider’s terms prohibit using those outputs to build competing models.
The source article lists prices of $10/$50 per million input/output tokens for Astra, compared with $2/$10 for Kimi K3.1 Pro and $0.30/$1.20 for DeepSeek’s model at peak hours. It also gives scores of 57, 46 and 32, respectively, on the same independent index. These are the article’s reported figures, not proof that one model is equivalent to another: the cheaper models scored lower on that index, and the source does not specify the index’s name or methodology in the supplied material.
For businesses, the allegation adds a reason to ask vendors about training-data provenance and contractual exposure when selecting models. The source article advises keeping a fallback and testing models on a company’s own work. For users, it reports no immediate change to the apps, while warning against unofficial, low-cost API resellers whose access and practices may be unclear.
Popular right now
De'Longhi Magnifica Evo Automatic Espresso Machine with Milk Frother – Silver
KitchenAid Artisan Series 5-Quart Tilt-Head Stand Mixer, Dried Rose (KSM150PS)
Shark Navigator Robot Vacuum with 30-Day Self-Emptying Base, White (AV2110S)
As an affiliate, we earn on qualifying purchases.
Distillation and Earlier Allegations
Distillation is a standard model-training technique. A “teacher” model generates responses to prompts, and a “student” model is trained to imitate them. AI labs commonly use the method to make smaller or cheaper models based on systems they control. The U.S. government advisory cited in the source material calls distillation a “legitimate and useful technique in AI research.” The controversy described here is not the technique by itself, but the alleged use of a competitor’s outputs, the scale of that use and whether access violated service rules.
The report follows earlier public allegations and disputes. The source article says OpenAI linked a July campaign to extract hidden reasoning from its models to people associated with Moonshot AI, and that U.S. security agencies named DeepSeek and Moonshot in an advisory. It also says China’s commerce ministry called the accusations baseless. These are positions and allegations attributed to the parties; they do not confirm the new claim about Astra. The supplied source material does not provide the advisory’s full text or the companies’ responses to the earlier claims.
OpenAI’s terms, as summarized by the source, bar using its outputs to build competing models. That makes the allegations potentially relevant to contract enforcement. Whether any law was broken would depend on facts not established in the public accounts, including how outputs were obtained and what agreements applied. The source’s discussion of possible fraud is an interpretation, not a finding by a court or regulator.
Evidence Behind the Astra Claim
The central allegation remains unconfirmed publicly. Reuters said OpenAI, DeepSeek and Moonshot did not immediately respond. The supplied reporting does not include the identities of The Information’s sources, documentary evidence, a technical analysis, or a response from either Chinese lab. It is also unclear whether the named models were trained directly on Astra outputs, how much material may have been involved, and whether any access was obtained through accounts that violated provider rules.
It is not clear whether the models’ performance or release materials can establish the origin of their training data. Similar capabilities can arise from different training methods, and claims about a model’s provenance require evidence beyond price comparisons, benchmark scores or the fact that distillation is technically possible. No legal determination or regulator finding about the alleged conduct is included in the supplied material.
Responses and Evidence to Watch
The next key developments would be direct responses from OpenAI, DeepSeek and Moonshot, along with any additional reporting that provides evidence about the alleged outputs, access methods or scale. The Information’s full account may contain details unavailable in its public summary; the source material does not establish what further evidence it includes.
Companies considering these models can review vendor terms and ask how training data is sourced, while testing systems against their own requirements. Any assessment should distinguish documented facts from vendor statements and outside allegations. For users, no immediate service change is reported; the practical status may change if a provider alters access, terms or product availability, but no such change is confirmed here.
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, naming DeepSeek-V4.1-Flash-1017 and Moonshot’s Kimi K3.1 Pro. The account cited people with direct knowledge.
Have the companies confirmed the report?
No. Reuters said OpenAI, DeepSeek and Moonshot did not immediately respond to requests for comment. The allegation has not been publicly confirmed in the supplied reporting.
Is distillation itself improper?
Not by itself. Distillation is a common training technique, including for building smaller models from a lab’s own systems. The disputed questions are whether a competitor’s outputs were used, at what scale, and whether the method or access breached applicable terms.
What evidence about the alleged training is public?
The supplied accounts do not state how many outputs were used, how they were obtained or what method was involved. They also do not provide technical evidence or a legal finding establishing a violation.
Does the report change anything for users now?
The source article reports no immediate change to the apps. It advises users to be cautious with unofficial, low-cost API resellers, but no new restriction or service change is confirmed.
Source: ThorstenMeyerAI.com
