🔍 Read the full analysis: Could Distillation Explain Similarities Between China’s AI Models And GPT-6 Astra? on ThorstenMeyerAI.com
TL;DR
The Information reported that DeepSeek and Moonshot AI relied heavily on outputs from OpenAI’s GPT-6 Astra to train recently released models, citing people with direct knowledge. The companies have not confirmed the allegation, and public reporting has not disclosed the scale or method. The central issue is whether rival-model distillation involved access or use prohibited by OpenAI’s terms.
The Information reported October 28 that Chinese AI labs DeepSeek and Moonshot AI relied heavily on outputs from OpenAI’s GPT-6 Astra to train models released this month. The allegation, attributed to people with direct knowledge, has not been confirmed by any of the three companies, leaving unanswered how the outputs were obtained and how extensively they shaped the models.
Reuters summarized the report as saying that Moonshot’s Kimi K3.1 Pro and DeepSeek-V4.1-Flash-1017 benefited from distilling Astra. Kimi K3.1 Pro was released October 23, while DeepSeek’s update launched October 17, according to the supplied source material. The Information’s public report described Astra as OpenAI’s most powerful released model and said Chinese AI labs were among a growing customer base seeking to extract its capabilities. The remainder of the article is paywalled.
OpenAI, DeepSeek and Moonshot did not immediately respond to Reuters, according to the report summary. No public evidence supplied with the allegation specifies the number of prompts involved, the training process, or whether the models’ results can be traced to Astra outputs. Moonshot’s launch post attributed Kimi K3.1 Pro’s gains to larger post-training compute and large-scale reinforcement learning. DeepSeek’s update note, as described in the source, did not identify its training data.
Separately, Korean broadcaster SBS reported October 29 that sources said Chinese labs had shifted from targeting Anthropic’s Claude to OpenAI models after Anthropic tightened defenses. SBS also quoted an OpenAI spokesperson as saying the company takes the issue seriously and continues working on it. The original account could not be independently checked against The Information’s paywalled report, so those details remain attributed to SBS rather than established facts.
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 focus attention on the boundary between ordinary model distillation and large-scale use of a competitor’s outputs. Distillation is a common training technique: a student model learns from responses generated by a larger teacher model. The technique itself is not evidence of misconduct; the dispute is about whose model supplied the outputs, how access was obtained, and whether the use complied with applicable terms.
The commercial stakes are clear in the supplied price and benchmark figures, though they do not establish how the models were trained. The source lists Astra at $10 per million input tokens and $50 per million output tokens, compared with Kimi K3.1 Pro at $2 and $10, and DeepSeek at $0.30 and $1.20 during peak hours. On the same independent index cited by the source, the models score 57, 46 and 32, respectively. Those figures indicate different listed prices and scores, not equivalent capabilities or proof that distillation produced any particular result.
For organizations choosing AI systems, the report is a reason to ask vendors about training-data provenance and evaluation, while testing models on their own workloads. The source also advises keeping a fallback model and avoiding low-cost, unofficial API resellers. It does not report a change to the public apps or services, and it provides no basis for concluding that users need to take immediate action.
Popular right now
Breville Bambino Espresso Machine, Brushed Stainless Steel (BES450BSS)
Dyson V8 Cyclone Cordless Vacuum
KitchenAid Classic Series 4.5 Quart Tilt-Head Stand Mixer K45SS, Onyx Black
As an affiliate, we earn on qualifying purchases.
Earlier Disputes Over Model Extraction
The allegation follows earlier disputes about attempts to extract model capabilities. The source material says OpenAI linked a July campaign to obtain hidden reasoning from its models to individuals associated with Moonshot AI in a September account. It also says U.S. security agencies named DeepSeek and Moonshot in an advisory. Those are claims and official statements described by the source, not independent proof that either lab used Astra outputs for the models at issue.
The same account says China’s commerce ministry called the accusations baseless. The source also describes a U.S. government advisory as recognizing distillation as a legitimate and useful research technique, while warning of problematic conduct involving rival models and access through fake or resold accounts. The distinction matters: a familiar training method can be used in different ways, and the relevant facts here have not been made public.
OpenAI’s terms, according to the source, prohibit using its outputs to build competing models. Whether any such terms were breached would depend on the agreement and evidence about the access and training process. That is separate from whether the conduct could involve deception in obtaining access; the available reporting does not establish either conclusion.
„DeepSeek and Moonshot “relied heavily on outputs from OpenAI’s powerful GPT-6 Astra model to train their latest models.”“
— The Information, as summarized by Reuters
The Evidence Behind the Allegation
The public account rests on unnamed sources cited by The Information, whose full report is behind a paywall. The supplied material includes no prompt logs, training records, account information, or technical comparison showing how much Astra output entered either model’s training. It is also not clear whether the reported use involved direct API access, third-party accounts, or another route.
Neither DeepSeek nor Moonshot has publicly confirmed the claim in the cited coverage, and Reuters reported that the companies and OpenAI did not immediately respond. The companies’ launch materials describe other training approaches or omit training-data details, but that does not by itself confirm or disprove the allegation. No public enforcement action or finding of a contract breach is identified in the source material.
The figures comparing prices and index scores are also not a test of training provenance. The source does not provide the index’s full methodology or establish that the reported distillation caused any performance differences. Further reporting or disclosures would be needed to resolve those questions.
Responses and Technical Evidence to Watch
The next developments to watch are public responses from OpenAI, DeepSeek and Moonshot, along with any further reporting that provides detail about the alleged data collection and training. Technical evidence, such as documented access patterns or independently examined training records, could help clarify whether Astra outputs were used and at what scale.
Any claim of a terms violation would also depend on the applicable contract and how access was obtained. The source material does not identify a lawsuit, regulator decision, or formal investigation arising from this report, so those outcomes should not be presumed. Until more evidence is released, the reported use remains an allegation rather than a confirmed account.
For AI buyers, the practical next step is routine vendor diligence: ask how models are trained, compare results on relevant tasks, and maintain a backup option. Those measures are prudent procurement practices, not a response to a verified disruption. The source reports no immediate change for users of the models’ apps.
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, citing people with direct knowledge. The allegation has not been confirmed by the companies.
Which models were named?
The report named Moonshot’s Kimi K3.1 Pro and DeepSeek-V4.1-Flash-1017, according to Reuters’ summary of the story.
Is distillation itself prohibited?
No general conclusion follows from the reporting. Distillation is a standard training technique; the dispute is whether a rival’s outputs were used in a way that breached the provider’s terms or involved improper access. The available source material does not establish that either happened.
Have OpenAI, DeepSeek or Moonshot confirmed the claim?
No. Reuters reported that none of the three companies immediately responded to its request for comment. The supplied material contains no direct confirmation from DeepSeek or Moonshot.
Does the report change anything for users now?
The source reports no immediate change to the apps. Users and businesses can continue evaluating models on their own needs, while avoiding unofficial API resellers and checking vendor information where available.
Source: ThorstenMeyerAI.com
