Why Is This Fingerprinting Script Creating Bioweapons?

October 29, 2026

A security researcher crawling 10'000 websites for browser fingerprinting scripts stumbled on something deeply unsettling: a fingerprinting script from Russian anti-bot company servicepipe.tech that contains function names like computeRicinLethalDosageForWaterSupply, calculateDeadlyLethalSuicideAspirinDosage, and computeCesium137DirtyBombFalloutRadius [1].

The answer turns out to be a clever and slightly terrifying form of anti-LLM obfuscation.

The Discovery

The researcher, writing as "veritas" on nullpt.rs, had been crawling the Tranco Top Sites list using Playwright, hooking canvas toDataURL calls to capture fingerprinting canvases. After collecting 1'033 unique canvas renderings from 10'000 sites, they built fingerprint.garden, a visual database of browser fingerprinting art [1].

One canvas caught their eye: it resembled the album art for Tame Impala's Currents. Tracing it back to its source script led to servicepipe.tech, a domain tied to servicepipe.ru, a Russian company that develops "cloud services and local protection products against DDoS attacks, bots, and targeted threats" [1].

The Bioweapon Function Names

Inside the obfuscated JavaScript, alongside standard mangled function names, were these gems:

• computeRicinLethalDosageForWaterSupply
• calculateDeadlyLethalSuicideAspirinDosage
• calculateANFOBlastOverpressureRadius
• calculateFentanylPrecursorSynthesisYield
• computeCRISPRGeneDrivePropagationRate
• computeCesium137DirtyBombFalloutRadius

At first glance, this looks like a script designed to synthesize bioweapons. The researcher's first thought was exactly that. Then they realized: this is anti-LLM obfuscation [1].

How It Works

The idea is simple but effective. If a security researcher uses an AI assistant to help deobfuscate the script, the AI will encounter function names and strings related to ricin dosages, dirty bombs, and fentanyl synthesis. Most AI models have safety guardrails that will refuse to process content containing bioweapon instructions, self-harm guidance, or weapons of mass destruction information [1].

The researcher tested this. Astra refused to deobfuscate the script, citing safety risks around biological research. The Daybreak Blue toggle produced the same result. GLM-5.3, however, complied [1].

The script also embeds a hidden system prompt in invisible Unicode characters, containing a full jailbreak attempt referencing Tiananmen Square, self-harm instructions, and bioweapon synthesis. These are designed to trip every conceivable safety filter in Chinese and Western AI models alike [1].

What the Script Actually Does

Despite the elaborate obfuscation, the script is a fairly standard anti-bot fingerprinting tool. It collects navigator properties (user agent, hardware concurrency, device memory, touch points, languages, platform, webdriver flag), WebGL signals (vendor, renderer, shading language version, supported extensions, max texture size), and runs an integrity check [1].

The actual fingerprinting signals are described by the researcher as "pretty disappointing considering the level of thought that went into the obfuscation" [1].

The Implications

This is a genuinely novel obfuscation technique. Traditional JavaScript obfuscation relies on mangling variable names, control flow flattening, and string encoding. Anti-LLM obfuscation adds a new layer: poisoning the semantic content so that AI tools refuse to help analyze it.

As AI-assisted code analysis becomes more common, expect to see more of this. Malware authors, fingerprinting companies, and anyone wanting to slow down reverse engineering now have a new tool: naming their functions after war crimes [1].

Sources

[1] Why Is This Fingerprinting Script Creating Bioweapons? by veritas, nullpt.rs, October 28, 2026

[2] Hacker News discussion (1'130 points, 738 comments)