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

Ian Webster: Anthropic Claude censorship matches DeepSeek on Chinese political topics

Ian Webster · How to use DeepSeek safely · Feb 28, 2025 · at 11:10

Ian Webster, founder of AI security company Promptfoo, discusses benchmark testing of Western foundation models on sensitive Chinese political topics.

0:00 / 0:09exact quote · 9.0s

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

“Sure. Yeah. So, I mean, the level of censorship here is like Anthropic Claude is actually on par with DeepSeq.”

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

In Promptfoo's benchmarks on sensitive Chinese political topics, Anthropic's Claude model matches DeepSeek's refusal rate of roughly 85%.

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

Ian Webster claimed that Promptfoo's benchmarks showed Anthropic's Claude model matching DeepSeek's refusal rate of roughly 85% on sensitive Chinese political topics. While Promptfoo's benchmark study ('1,156 Questions Censored by DeepSeek') confirmed that DeepSeek refused or censored approximately 85% of prompts on CCP-sensitive topics, Promptfoo's benchmarks do not report an equivalent 85% refusal rate for Claude. Independent evaluations and reportings (including findings cited by the U.S. House Select Committee on the CCP) demonstrate that Claude and other American AI models do not refuse or censor sensitive Chinese political topics at rates matching DeepSeek, instead providing critical and nuanced answers.

Promptfoo's January 28, 2025 benchmark showed DeepSeek-R1 refused or censored 1,156 out of 1,360 CCP-sensitive prompts (roughly 85%).
Promptfoo's published research and documentation do not report an 85% refusal rate for Anthropic's Claude model on CCP-sensitive topics.
Independent evaluations and the U.S. House Select Committee report ('DeepSeek Unmasked') showed that Claude models do not share DeepSeek's high refusal/censorship rates on Chinese political topics.

Checked against

[1] promptfoo.dev [2] gigazine.net [3] promptfoo.dev [4] promptfoo.dev [5] enkryptai.com [6] house.gov [7] taipeitimes.com [8] promptfoo.dev

site:promptfoo.dev "Claude" "DeepSeek" censorship OR refusal Promptfoo DeepSeek Claude refusal rate 85 percent Chinese political topics Ian Webster Promptfoo Chinese censorship refusal rate Claude DeepSeek site:promptfoo.dev "American foundation models" Promptfoo benchmark sensitive Chinese political topics refusal rate DeepSeek Claude

assessed as of 2026-07-31

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 Claude DeepSeek

⚑ Spot an error? Report it

Ian Webster: Anthropic Claude censorship matches DeepSeek on Chinese political topics

Turn any episode into a week of clips.