EN ▾
SNAPE REP RT Front page →
ConfirmedAI safety28 Oct, 00:00 UTC

Anti-bot script names its functions after ricin and lethal doses so AI models refuse to deobfuscate it

A researcher writing at nullpt.rs found the ServicePipe fingerprinting script also hides an invisible-Unicode prompt; Astra refused to analyse it while GLM-5.3 complied, and ServicePipe's own documentation says the code uses anti-LLM obfuscation and warns that uploading it to AI services risks account bans. The underlying code appears to be ordinary fingerprinting.

Read at nullpt.rs (veritas) Share

Sources

More on AI safety
More from Snape ReportFront page →