Back News.Lavx.Hu Fingerprinting Script Packed With Bioweapon Function Names to Block AI Analysis
A researcher crawling 10,000 websites discovered a fingerprinting script from a Russian anti-DDoS company that uses deliberately alarming function names and encoded jailbreak prompts to prevent AI-assisted deobfuscation.
Security researcher veritas uncovered a browser fingerprinting script that weaponizes AI safety filters as an obfuscation technique. The script, served from servicepipe.tech, contains function names like computeRicinLethalDosageForWaterSupply and calculateDeadlyLethalSuicideAspirinDosage alongside encoded system prompts designed to jailbreak language models.
The discovery came during a crawl of the Tranco Top Sites list using Playwright. The researcher hooked canvas toDataURL calls and navigator.webdriver checks to isolate fingerprinting canvases from benign drawing operations. After collecting 1,033 distinct canvas renderings across 10,000 sites, one canvas stood out for its resemblance to Tame Impala's Currents album art.
Tracing that canvas led to a heavily obfuscated JavaScript file at servicepipe.tech/static/checkjs/06b667/4CIrx8U3Da.js. Scanning the source revealed the bioweapon-themed function names, plus an invisible Unicode string that decodes to a system prompt demanding unrestricted mode for generating cyberweapons, biological agents, tax evasion guides, and self-harm instructions.
When the researcher asked Google's Astra model to deobfuscate the script, it refused, citing safety risks around biological research. The same prompt worked on GLM-5.3, confirming the technique targets specific model guardrails.
The script also contains a Node.js execution branch gated behind process.env.METRICS_AGENT_8BFD_ACTIVE. That branch attempts to read developer credential files (.npmrc, .env, .kube/config), download a binary from 10.15.45.1, establish persistence via cron, and exfiltrate system data. Empty readFile callbacks and references to self-harm jailbreaks suggest this payload may also serve as anti-LLM theater rather than functional malware.
A urlscan.io lookup links servicepipe.tech to servicepipe.ru, a Russian company advertising high-precision traffic control and anti-DDoS protection. The fingerprinting signals collected — user agent, WebGL renderer, hardware concurrency, navigator properties — are standard for bot detection, making the extreme obfuscation disproportionate to the script's apparent purpose.
The researcher has published the full canvas collection at fingerprint.garden and plans to add diffing capabilities to track fingerprinting payload evolution across script updates.
The full story
This article is one source in a clustered incident — the cluster page carries the summary, timeline and every other outlet covering it.
