🔍 Read the full analysis: The Limits Of The Report On China’s New AI Models And GPT-6 Astra on ThorstenMeyerAI.com
Get the latest gadgets 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 outputs from OpenAI’s GPT-6 Astra to train recent models. Reuters said the three companies did not immediately respond; the public record lacks details on the volume, method and source of the outputs. Distillation is a common training technique, but using a rival’s model may breach its terms.
The Information reported Oct. 28 that Chinese AI labs DeepSeek and Moonshot AI relied heavily on outputs from OpenAI’s GPT-6 Astra to train recent models. The allegation, also reported by Reuters, has not been publicly confirmed by any of the three companies, and key details about how the outputs were collected remain unknown.
The report names Moonshot’s Kimi K3.1 Pro, released Oct. 23, and DeepSeek-V4.1-Flash-1017, released Oct. 17. Reuters summarized The Information’s account as saying both models “benefitted from distilling Astra.” The Information’s publicly available opening says Chinese AI labs were among a growing group of customers seeking to extract the model’s capabilities; the rest of its report is behind a paywall.
Reuters said OpenAI, DeepSeek and Moonshot did not immediately respond to its requests for comment. The source article also cites a Korean broadcaster’s summary of the report, which says an OpenAI spokesperson described the issue as serious. That wording has not been independently checked against The Information’s full article in the supplied material, so it does not amount to a company confirmation of the specific allegations.
The public reporting supplies no figures for the number of prompts or outputs, no description of the alleged collection method, and no evidence showing how much any borrowed outputs contributed to either model. Moonshot’s launch post attributed Kimi K3.1 Pro’s gains to more post-training compute and large-scale reinforcement learning; DeepSeek’s update note, as described in the source, did not discuss training data. Those company statements do not resolve the allegations.
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 Provenance Matters
The issue matters because model outputs can carry valuable capabilities, including response patterns and, where exposed, reasoning traces. Training a model on another system’s answers can reduce the work needed to reproduce some of its performance. The report raises questions about whether that happened at scale, but the evidence made public so far does not establish the extent or impact.
There is also a practical difference between distillation as a technique and the way a lab obtains a rival’s outputs. Providers commonly use outputs from their own larger models to train smaller, less expensive systems. If a company uses a competitor’s service in ways prohibited by its terms, that may create a contractual dispute; allegations of fake or resold accounts could raise separate questions about how access was obtained. Whether any law was broken depends on facts and legal analysis not supplied by the report.
For businesses choosing AI systems, the allegation is a reminder to ask vendors about training-data provenance and access controls, and to test models on their own tasks rather than relying only on published scores. The source material gives a price and index comparison for Astra and the two Chinese models, but those figures do not establish that one model’s results came from another. They should not be treated as proof of copying or equivalence.
Products Worth Considering
Bose QuietComfort Wireless Bluetooth Headphones with Active Noise Cancellation, Mic, USB‑C Charging, Deep Bass, 24‑Hour Battery – White Smok
DJI Osmo Pocket 3 Creator Combo Vlogging Camera with Mic 2, 1 CMOS, 4K/120fps, 3-Axis Stabilization, 128GB Card
Sony WH-1000XM6 Wireless Noise-Canceling Over-Ear Headphones, Blue, Bundle
As an affiliate, we earn on qualifying purchases.
Distillation and Earlier Allegations
Distillation is a standard model-training approach: a larger “teacher” model generates answers to prompts, and a smaller “student” model is trained to imitate them. A U.S. government advisory cited in the source describes distillation as a legitimate and useful research technique. The central dispute in this case is not the method alone, but the allegation that labs used a rival’s outputs at scale and potentially in ways that violated service terms.
The report follows earlier public accusations. The source says OpenAI tied a “core cluster” of a July campaign to extract hidden reasoning from its models to people associated with Moonshot AI in a September account. It also says U.S. security agencies named DeepSeek and Moonshot in an advisory, while China’s commerce ministry called the accusations baseless. These are claims and responses from the parties involved, not independent proof of the new training allegation.
The source also describes an earlier Anthropic account alleging large numbers of exchanges with Claude by several labs. That history provides context for concerns about competitors extracting model capabilities, but it does not independently verify what happened with GPT-6 Astra or establish that the new models were trained using its outputs.
“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
Evidence Still Missing
The core allegation remains unverified in public by the companies involved. The available account does not say how many Astra outputs were used, which prompts were sent, who obtained access, whether accounts were shared or resold, or how the alleged data was incorporated into training. The Information’s full report is paywalled, limiting independent review of its sourcing and supporting details.
It is also unclear whether any use of Astra outputs violated OpenAI’s terms, whether access was obtained through misrepresentation, or whether the models’ performance can be linked to those outputs rather than to the training methods their developers describe. No public response from DeepSeek or Moonshot addressing the specific claims is included in the source material.
Responses and Evidence to Watch
The next meaningful development would be a detailed response from OpenAI, DeepSeek or Moonshot, or the release of evidence clarifying the alleged data collection and training process. Any assessment of contract or legal consequences would also depend on the applicable terms, access records and evidence of use, none of which is publicly set out in the material provided.
For now, companies evaluating these systems can review vendor disclosures, ask how training data was sourced, and keep alternative providers available. Users have no reported need to change how they use the official apps based on this allegation alone. The claim should remain framed as a report—not an established finding—unless further evidence or confirmation emerges.
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 Kimi K3.1 Pro and DeepSeek-V4.1-Flash-1017.
Have the companies confirmed the report?
No confirmation is included in the available reporting. Reuters said OpenAI, DeepSeek and Moonshot did not immediately respond to its requests for comment.
Is distillation itself improper?
Not necessarily. Distillation is a common technique for training smaller models from a larger model’s outputs. The questions here concern whether a competitor’s outputs were used, how they were obtained, and whether that use violated service terms or other rules.
What evidence has been made public about the alleged training?
The source material identifies the models and summarizes The Information’s allegation, but provides no prompt counts, collection method or measurement of the outputs’ effect on either model.
Source: ThorstenMeyerAI.com
