Wikipedia's Guide to Spotting AI Is Now Being Used To Hide AI
(Thursday January 22, 2026 @05:00AM (BeauHD)
from the full-circle dept.)
- Reference: 0180635714
- News link: https://news.slashdot.org/story/26/01/22/015250/wikipedias-guide-to-spotting-ai-is-now-being-used-to-hide-ai
- Source link:
Ars Technica's Benj Edwards reports:
> On Saturday, tech entrepreneur Siqi Chen [1]released an open source plugin for Anthropic's Claude Code AI assistant that instructs the AI model to stop writing like an AI model. Called "Humanizer," the simple prompt plugin feeds Claude a list of 24 language and formatting patterns that Wikipedia editors have [2]listed as chatbot giveaways. Chen published the plugin on GitHub, where it has picked up over 1,600 stars as of Monday. "It's really handy that Wikipedia went and collated a detailed list of 'signs of AI writing,'" Chen [3]wrote on X. "So much so that you can just tell your LLM to... not do that."
>
> The source material is a guide from WikiProject AI Cleanup, a group of Wikipedia editors who have been hunting AI-generated articles since late 2023. French Wikipedia editor Ilyas Lebleu founded the project. The volunteers have tagged over 500 articles for review and, in August 2025, [4]published a formal list of the patterns they kept seeing.
>
> Chen's tool is a " [5]skill file " for Claude Code, Anthropic's terminal-based coding assistant, which involves a Markdown-formatted file that adds a list of written instructions (you can [6]see them here ) appended to the prompt fed into the large language model (LLM) that powers the assistant. Unlike a normal system prompt, for example, the skill information is formatted in a standardized way that Claude models are fine-tuned to interpret with more precision than a plain system prompt. (Custom skills require a paid Claude subscription with code execution turned on.)
>
> But as with all AI prompts, language models don't always perfectly follow skill files, so does the Humanizer actually work? In our limited testing, Chen's skill file made the AI agent's output sound less precise and more casual, but it could have some drawbacks: it won't improve factuality and might harm coding ability. [...] Even with its drawbacks, it's ironic that one of the web's most referenced rule sets for detecting AI-assisted writing [7]may help some people subvert it .
[1] https://github.com/blader/humanizer
[2] https://en.wikipedia.org/wiki/Wikipedia:Signs_of_AI_writing
[3] https://x.com/blader/status/2013015738622284156
[4] https://en.wikipedia.org/wiki/Wikipedia:Signs_of_AI_writing
[5] https://support.claude.com/en/articles/12512198-how-to-create-custom-skills
[6] https://github.com/blader/humanizer/blob/main/SKILL.md
[7] https://arstechnica.com/ai/2026/01/new-ai-plugin-uses-wikipedias-ai-writing-detection-rules-to-help-it-sound-human/
> On Saturday, tech entrepreneur Siqi Chen [1]released an open source plugin for Anthropic's Claude Code AI assistant that instructs the AI model to stop writing like an AI model. Called "Humanizer," the simple prompt plugin feeds Claude a list of 24 language and formatting patterns that Wikipedia editors have [2]listed as chatbot giveaways. Chen published the plugin on GitHub, where it has picked up over 1,600 stars as of Monday. "It's really handy that Wikipedia went and collated a detailed list of 'signs of AI writing,'" Chen [3]wrote on X. "So much so that you can just tell your LLM to... not do that."
>
> The source material is a guide from WikiProject AI Cleanup, a group of Wikipedia editors who have been hunting AI-generated articles since late 2023. French Wikipedia editor Ilyas Lebleu founded the project. The volunteers have tagged over 500 articles for review and, in August 2025, [4]published a formal list of the patterns they kept seeing.
>
> Chen's tool is a " [5]skill file " for Claude Code, Anthropic's terminal-based coding assistant, which involves a Markdown-formatted file that adds a list of written instructions (you can [6]see them here ) appended to the prompt fed into the large language model (LLM) that powers the assistant. Unlike a normal system prompt, for example, the skill information is formatted in a standardized way that Claude models are fine-tuned to interpret with more precision than a plain system prompt. (Custom skills require a paid Claude subscription with code execution turned on.)
>
> But as with all AI prompts, language models don't always perfectly follow skill files, so does the Humanizer actually work? In our limited testing, Chen's skill file made the AI agent's output sound less precise and more casual, but it could have some drawbacks: it won't improve factuality and might harm coding ability. [...] Even with its drawbacks, it's ironic that one of the web's most referenced rule sets for detecting AI-assisted writing [7]may help some people subvert it .
[1] https://github.com/blader/humanizer
[2] https://en.wikipedia.org/wiki/Wikipedia:Signs_of_AI_writing
[3] https://x.com/blader/status/2013015738622284156
[4] https://en.wikipedia.org/wiki/Wikipedia:Signs_of_AI_writing
[5] https://support.claude.com/en/articles/12512198-how-to-create-custom-skills
[6] https://github.com/blader/humanizer/blob/main/SKILL.md
[7] https://arstechnica.com/ai/2026/01/new-ai-plugin-uses-wikipedias-ai-writing-detection-rules-to-help-it-sound-human/
Don't stuff beans up your nose (Score:2)
by xack ( 5304745 )
[1]Wikipedia's own "policy" [wikipedia.org] btw, by telling an AI not to be an AI, it's going to follow the advice to "not be an AI". AI is now emulating the behaviour of a naughty child.
[1] https://en.wikipedia.org/wiki/Wikipedia:Don't_stuff_beans_up_your_nose
Arms race (Score:2)
I think AI is going to be the end of the open web. There is already an arms race between slop makers and legitimate content curators, but the odds are in favor of the former - they have incentive, time and automated tools that can generate an endless pile of sometimes believable junk at their disposal. On the other side, limited human time and capital, and difficulty that will increase to distinguish between slop and actual content.
This will kill open collaboration on the internet, also killing most project