Assertion Partly supported AI assessment confidence: 90% certainty 4/5 debate potential 2/5

Matt Clifford: China Requires AI Developers to Submit Training Data to Government

Matt Clifford · Matt Clifford: The Bull & Bear Case for China's Ability to Challenge the US' AI Capabilities | E1172 · 20VC with Harry Stebbings · Jul 1, 2024 · at 18:42

Matt Clifford explains specific regulatory compliance mandates imposed by the Chinese government on large AI model developers.

0:00 / 0:17exact quote · 17.4s

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

“For example, if you train a large model, you have to, like, supply random samples of the training data to the to the CCP or to the government. You have to you have to, like, show it's, like, political, that it is not gonna undermine any of the, you know, state positions of the CCP on various issues.”

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

Under Chinese AI regulation, training a large model requires submitting random samples of training data to the government and proving the model does not contradict CCP state positions.

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

The claim contains two distinct assertions about Chinese AI regulations prior to July 2024. First, the claim that model developers must prove their models do not undermine CCP state positions is true: Article 4(1) of the CAC's 2023 Interim Measures and TC260 security standards require models to align with Socialist Core Values, undergo safety assessments, and refrain from generating content that subverts state power or endangers national security. Second, the claim that developers must submit random samples of training data to the government is partially inaccurate: TC260 standards mandate that developers perform internal random sampling on their own datasets (e.g., sampling 4,000 items per corpus) and submit self-assessment reports detailing data sources and rules to the CAC. However, the rules do not require routine submission of raw random data samples directly to government databases as a standard filing requirement.

Article 4(1) of the 2023 Interim Measures requires generative AI to adhere to Socialist Core Values and refrain from inciting subversion of state power.
TC260-003 standards require developers to perform internal data sampling (e.g., at least 4,000 items per training corpus) to check for illegal and negative information.
CAC algorithm filings require submitting self-assessment reports and dataset metadata, but do not mandate submitting raw random samples of training data directly to government repositories.

Checked against

[1] cambridge.org [2] aigovernance.com [3] pertamapartners.com [4] thomsonreuters.com [5] amazonaws.com [6] cyrilla.org [7] warontherocks.com [8] dig.watch

assessed as of 2026-07-23

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

China Chinese Communist Party

⚑ Spot an error? Report it

More from Matt Clifford

Prediction Not checkable as stated

Matt Clifford: The UK Can Return to Being the World's Richest Country

“I think the UK can go back to being pretty much the richest country in the world per capita. I really truly believe that.”

Matt Clifford Jul 1, 2024 ▶ 34:20 Matt Clifford: The Bull & Bear Case for China's Ability to Challenge the US' AI Capabilities | E1172 · 20VC with Harry Stebbings

Assertion Not checkable as stated

Matt Clifford: Value of Scaling Compute and Data for LLMs Is Flattening

“I think that we are seeing the flattening off of the value of just adding more compute and more data to language models.”

Matt Clifford Jul 1, 2024 ▶ 5:38 Matt Clifford: The Bull & Bear Case for China's Ability to Challenge the US' AI Capabilities | E1172 · 20VC with Harry Stebbings

Opinion

Matt Clifford: The Threat of Nuclear War Is Underrated

“I think nuclear war is really underrated.”

Matt Clifford Jul 1, 2024 ▶ 40:20 Matt Clifford: The Bull & Bear Case for China's Ability to Challenge the US' AI Capabilities | E1172 · 20VC with Harry Stebbings

Opinion

Matt Clifford: The EU AI Act Is a Mistake for AI Innovation

“I mean, I do think the EU AI Act is a mistake.”

Matt Clifford Jul 1, 2024 ▶ 31:49 Matt Clifford: The Bull & Bear Case for China's Ability to Challenge the US' AI Capabilities | E1172 · 20VC with Harry Stebbings

Assertion Partly supported

Matt Clifford: Local Councils Blocked US Tech Giants' UK Data Center Investments

“And the number one issue that these companies had, and I'm really not exaggerating, was we want to invest billions and billions of dollars to build compute infrastructure in the UK, and your local county councils in the middle of nowhere keep vetoing our data …”

Matt Clifford Jul 1, 2024 ▶ 35:14 Matt Clifford: The Bull & Bear Case for China's Ability to Challenge the US' AI Capabilities | E1172 · 20VC with Harry Stebbings

Opinion

Matt Clifford: Degrowth Is the Worst Acceptable Idea

“So I think probably like the worst idea in the world that it's acceptable to hold in polite company is like degrowth. Like the idea that like, oh, you know, like the real problem in the world is like Too many people, too much stuff, too much consumption, we sh…”

Matt Clifford Jul 1, 2024 ▶ 43:27 Matt Clifford: The Bull & Bear Case for China's Ability to Challenge the US' AI Capabilities | E1172 · 20VC with Harry Stebbings

Matt Clifford: China Requires AI Developers to Submit Training Data to Government

More from Matt Clifford

Turn any episode into a week of clips.