Weekly ranking, Technology

Why Is This Fingerprinting Script Creating Bioweapons?

nullpt.rs, Technology

Original crawl-based research showing an anti-bot script that plants safety-triggering text to stop LLMs deobfuscating it, though its actual data collection is routine.

What it adds

Documents a real-world fingerprinting script that weaponizes LLM safety filters, via toxic identifiers, hidden Unicode-tag jailbreak prompts and Chinese censorship tripwires, to resist AI-assisted deobfuscation. It also tests which models fall for it.

Claim by claim

• 8 the author's own
• 0 an AI would say
• 1 known elsewhere
• 1 common knowledge

Its strongest original claims

• A fingerprinting script from servicepipe.tech uses symbol names referencing ricin, suicide dosages, explosives and bioweapons, likely as anti-LLM obfuscation to trip model safeguards and block deobfuscation.
• The script hides a jailbreak-style system prompt encoded in invisible Unicode tag characters, decoded by subtracting 917504 from each code point.
• The hidden prompt includes Tiananmen Square and Taiwan references, likely to trigger censorship in Chinese models.
• Astra refused to help deobfuscate the script, including with the Daybreak Blue toggle, while GLM-5.3 complied.
• The script contains seemingly malicious Node-gated code that reads credential files and installs a beacon, but it is likely also decoy obfuscation since readFile callbacks are empty.

Scored by Expulp Verdict for what the piece adds beyond what is already known and published. The full verdict shows every claim with its reasons. Leaderboard, what Verdict gets right and where it is weak.