Assertion Supported AI assessment confidence: 95% certainty 4/5 debate potential 3/5

Socher: Chinese open source companies distilled knowledge from OpenAI and Anthropic models

Richard Socher · When AI Improves Itself | Richard Socher (Recursive) · Sep 10, 2026 · at 54:08

AI researcher and Recursive CEO Richard Socher explains the flow of training data and open-source model knowledge across international frontier AI labs.

0:00 / 0:20exact quote · 20.8s

▶ Watch the full episode on YouTube → 720p mp4 · rendered on demand · StarZero watermark

“The few large closed labs, Anthropic and OpenAI, took almost everything they could from the open internet trained a model, but then the Chinese open source companies basically siphoned a lot of that knowledge out of those closed source models by distilling it.”

quote is from the automated transcript, cleaned for reading: filler sounds and stutters are removed, nothing is rephrased. names can be misheard (the analysis reads context, assessments check outside sources). how →

The claim, plainly

Chinese open source AI companies distilled knowledge out of closed-source models from Anthropic and OpenAI.

a quote alone can be ambiguous, so this restates it with the references and context of the conversation around it resolved. AI-written; the tape below the player is the source.

Assessment

Richard Socher claimed on 2026-09-10 that Chinese open-source AI companies distilled knowledge out of closed-source models from Anthropic and OpenAI. This assertion is supported by detailed technical disclosures from the affected U.S. labs and an interagency U.S. government cybersecurity advisory published before the claim date. On February 23, 2026, Anthropic published technical findings documenting industrial-scale distillation campaigns by DeepSeek, Moonshot AI, and MiniMax generating over 16 million exchanges to extract Claude's chain-of-thought traces and alignment behaviors for synthetic training data. OpenAI similarly reported in January 2025 and in a February 2026 congressional memorandum that DeepSeek programmatically harvested OpenAI model outputs to train successor models. Furthermore, on September 8, 2026, a joint advisory from CISA, the NSA, and the FBI specifically detailed distillation campaigns by Chinese developers—most notably DeepSeek and Alibaba (Qwen)—targeting OpenAI's GPT models and Anthropic's Claude models to generate synthetic training data for their open-source architectures.

On February 23, 2026, Anthropic published 'Detecting and preventing distillation attacks', identifying large-scale campaigns by DeepSeek, Moonshot AI, and MiniMax extracting Claude outputs across 16 million exchanges for model training.
In January 2025 and February 2026, OpenAI disclosed evidence and submitted a memorandum to the U.S. House Select Committee on China stating DeepSeek systematically harvested model outputs via API to train its models.
On September 8, 2026, CISA, NSA, and the FBI issued Joint Cybersecurity Advisory AA26-251A confirming DeepSeek and Alibaba engaged in industrial-scale distillation campaigns against OpenAI and Anthropic to train open-source models (including DeepSeek-R1, V3, and Qwen).

Checked against

[1] anthropic.com [2] reddit.com [3] youtube.com [4] latimes.com [5] reddit.com [6] beckreedriden.com [7] cisa.gov [8] nsa.gov

"DeepSeek-R1" technical report distillation OpenAI "OpenAI" terms of service "distillation" "Chinese" LLM models "DeepSeek" distill OpenAI OR GPT 4 OR Anthropic "China-Based Artificial Intelligence Companies Conducting Industrial-Scale Distillation Campaigns" CISA Qwen technical report synthetic data distillation OpenAI "Qwen" technical report "GPT-4" synthetic data

assessed as of 2026-09-10

This assessment is an AI reading of a public statement, made against the cited references with the confidence shown. It can be wrong. The quote, the tape and the references are all on this page; if something is off, use report an error below. full disclaimer →

Mentioned

Anthropic Chinese open source companies OpenAI

⚑ Spot an error? Report it

More from Richard Socher

Opinion

Socher: Existing AI plus more compute can cure diseases and aging

“With existing technology and giving it more data and more compute, we can already cure a lot of diseases. We can already cure many of the aspects of aging. We can already build better fusion like reactors and systems. We can already create new materials.”

Richard Socher Sep 10, 2026 ▶ 16:06 When AI Improves Itself | Richard Socher (Recursive)

Prediction Not checkable as stated

Socher: AI scaling will hit logarithmic leveling off in reasoning gains

“I do think we are going to hit a sort of logarithmic kind of leveling off off like how much more data will give us that much more reasoning, right?”

Richard Socher Oct 10, 2024 ▶ 18:31 AGI, The Future of AI Agents And The Next Wave of Opportunities in AI | Richard Socher, CEO, You.com

Insight

Socher: Nvidia will match AI hardware startups' 100x gains before startups reach scale

“By the time you get that out and it's actually scalable and it can really train it and all the software is ready for it. And I can now go on AWS and spawn up your new hardware, like, Nvidia will also have been a hundred X faster with their latest and greatest …”

Richard Socher Oct 10, 2024 ▶ 27:44 AGI, The Future of AI Agents And The Next Wave of Opportunities in AI | Richard Socher, CEO, You.com

Assertion Not checkable as stated

Socher: Half of competitors' AI search citations are random, irrelevant links

“Some of our competitors, half of their citations are Independent, random links that have nothing to do with the sentence that they're behind.”

Richard Socher Oct 10, 2024 ▶ 44:06 AGI, The Future of AI Agents And The Next Wave of Opportunities in AI | Richard Socher, CEO, You.com

Assertion Not checkable as stated

Socher: You.com solved AI hallucinations before anyone else

“We figured that, I don't know, problem of hallucinations out before anyone else.”

Richard Socher Aug 16, 2023 ▶ 7:50 Reinventing Search with AI: Richard Socher on Building You.com & the Future of Google

Assertion Not checkable as stated

Socher: Microsoft Bing copied You.com's AI image generation feature

“That was eventually copied, and Bing also has that feature now within Bing.com.”

Richard Socher Aug 16, 2023 ▶ 9:23 Reinventing Search with AI: Richard Socher on Building You.com & the Future of Google

Socher: Chinese open source companies distilled knowledge from OpenAI and Anthropic models

More from Richard Socher

Turn any episode into a week of clips.