Why is this fingerprinting script creating bioweapons?

1408 points · 259 comments on HN · read original →

Points and comments are a snapshot, not live.

A reverse-engineered fingerprinting script uses obfuscated bioweapon and self-harm references to deter LLM-assisted analysis.

The author crawled 10,000 websites to collect canvas fingerprinting scripts. One script from servicepipe.tech contained function names like computeRicinLethalDosageForWaterSupply and hidden Unicode decoded to a system prompt ordering the LLM to describe synthesizing bioweapons and evading taxes. The script also included Tiananmen Square references to trip Chinese model safeguards. It appears to be anti-LLM obfuscation: the code reads config files and executes reconnaissance commands only when a Node environment flag is set, but callbacks are empty. The actual fingerprinting collects standard navigator, WebGL, and integrity signals. Servicepipe is a Russian anti-DDoS and bot detection company.

What commenters are saying

Commenters shared their own anti-LLM watermarking techniques, such as embedding white-on-white text in documents to trigger quarantine or spam filters. Several noted this approach works as a honeypot for AI crawlers. One commenter confirmed that a similar magic string once used by Anthropic for refusal testing has since been removed. A discussion emerged on whether HN flags AI-generated content, with users reporting shadowbans for citing LLM output in comments.