🔍 Read the full analysis: A Report Raises Questions About GPT-6 Astra And China’s New AI Models 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 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 respond to its requests for comment; the public report does not disclose the methods or scale of the alleged use.
The Information reported on October 28 that Chinese AI companies DeepSeek and Moonshot AI relied heavily on outputs from OpenAI’s GPT-6 Astra to train recent models, including versions released this month. Reuters summarized the report, but said OpenAI, DeepSeek and Moonshot did not immediately respond to its requests for comment; the allegation has not been publicly confirmed by the companies.
The Information’s report, by Jing Yang, Juro Osawa and Qianer Liu, cited people with direct knowledge, according to Reuters. It named Moonshot’s Kimi K3.1 Pro, released October 23, and DeepSeek-V4.1-Flash-1017, launched October 17, as models that benefited from distilling Astra. The Information’s public opening described Chinese AI labs as a growing customer base for Astra; the rest of its report is behind a paywall.
The available public account does not specify how many Astra outputs were used, what prompts or access methods were involved, or how much they contributed to either model’s capabilities. 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 its training data. Neither statement independently verifies or disproves the reported use of Astra outputs.
Reuters reported that the three companies did not immediately respond to its inquiries. A Korean broadcaster, SBS, separately summarized the report on October 29 and quoted an OpenAI spokesperson saying the company takes the issue very seriously and continues working on it. That detail is relayed through SBS, rather than independently confirmed here against The Information’s full report. The central allegation therefore remains a reported claim, not a finding established in the public record.
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 could sharpen a dispute over how companies may use a rival’s model outputs to train competing systems. Distillation is widely used to make models smaller or less expensive to run. The contested issue is not the technique by itself, but whether a company used another provider’s outputs at scale and in a way barred by its access terms.
The commercial stakes include both model costs and competitive position. The source article 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 reports scores of 57, 46 and 32, respectively, on the same independent index. Those figures are reported by the source and do not show that one model is interchangeable with another; the cheaper models score lower on that index.
For customers, the allegation may make training-data provenance and vendor disclosures more relevant when evaluating AI systems. It does not establish that users of the named models face an immediate change in service, nor does the report itself prove a legal violation. Businesses may want to understand how a model was trained and what contractual assurances a provider offers, while recognizing that public details remain limited.
Products Worth Considering
Apple MacBook Pro 16.2-inch with M5 Pro, 18-core CPU, 20-core GPU, 48GB Memory, 1TB SSD, Wi-Fi 7, Space Black
Apple MacBook Pro 16-inch (2021) – M1 Pro Chip, 10-Core CPU, 16-Core GPU, 16GB RAM, 1TB SSD, Space Gray
LG UltraGear 32GX850A-B 32-Inch 4K UHD Glossy OLED Gaming Monitor
As an affiliate, we earn on qualifying purchases.
Distillation and Earlier Allegations
Distillation is a training method in which a student model learns from responses produced by a larger teacher model. AI labs commonly use it with their own systems to create smaller, more economical models. The U.S. government advisory described in the source material calls distillation a legitimate and useful research technique. The controversy arises when the teacher model belongs to a competitor and its outputs are allegedly collected at scale in breach of access rules.
The report follows earlier public disputes about attempts to extract model capabilities. The source material says OpenAI linked a July campaign seeking 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 separate claims and responses; they do not establish the specific training practices alleged in the October report.
OpenAI’s terms, as described in the source article, prohibit using its outputs to build competing models. Whether that prohibition applies to particular conduct would depend on evidence about access, use and the relevant agreements. A possible terms-of-service breach is not automatically a proven crime; the source article frames the issue chiefly as contractual, with possible fraud questions if access was obtained through deceptive accounts.
“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 and Methods Still Missing
The public information does not identify the volume of outputs, collection methods, accounts used or specific training records behind the allegation. It also does not establish whether the named labs accessed Astra directly, obtained outputs through intermediaries, or used any outputs in a way that violated a particular agreement. The Information’s detailed report is paywalled, and the public summary does not supply supporting documents or data.
There is also no public response in the material from DeepSeek or Moonshot addressing the allegation directly, and OpenAI’s reported response is limited. It remains unclear whether the companies will issue fuller statements or whether further evidence will be made public. The account should be treated as an allegation attributed to sources, not a settled account of how either model was trained.
Company Responses and Evidence to Watch
The next developments to watch are direct responses from OpenAI, DeepSeek and Moonshot, along with any evidence about the scale and source of the model outputs. Further reporting, public documentation or an action tied to service terms could clarify what happened, but none is established in the material currently available.
For AI buyers, the immediate practical step is due diligence rather than assuming the allegation is proven: ask vendors about data provenance, training processes and contractual commitments, and test models on the tasks that matter to the organization. The source article also advises companies to keep a fallback and cautions users against cheap, unofficial API resellers. There is no reported change to the named consumer apps in the source material. AI services and their outputs carry limitations, and customers should not treat this report as proof that any particular model is unsafe or unlawful.
Key Questions
What does the report allege?
The Information reported that DeepSeek and Moonshot AI relied heavily on OpenAI’s 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 publicly confirmed the allegation in the source material.
What is AI model distillation?
Distillation trains a smaller student model using responses from a larger teacher model. Labs use the technique for their own systems; the dispute in this report concerns the alleged use of a competitor’s outputs and whether that use breached access terms.
Has OpenAI confirmed that the models were trained on Astra?
No public confirmation is provided. Reuters said OpenAI, DeepSeek and Moonshot did not immediately respond to its requests. SBS later quoted an OpenAI spokesperson saying the company takes the issue seriously, but that is not confirmation of the specific allegation.
Does the report prove that a law was broken?
No. The public account does not establish how outputs were obtained or used, and it does not present a legal finding. The source article says OpenAI’s terms prohibit using its outputs to build competing models, but whether a breach occurred would depend on facts and applicable agreements.
Do users need to change how they use these models now?
The source material reports no immediate change to the named apps. The allegation remains unconfirmed. Organizations evaluating AI vendors can ask about training-data provenance and test services on their own tasks; users should be cautious with unofficial API resellers.
Source: ThorstenMeyerAI.com
