What is Villain's Hat?

Villain's Hat is a pre-ship misuse audit for AI product teams too small to hire a Trust and Safety function. It uses the VRSD methodology to identify the anti-ICP: users you never designed for who exploit your product in ways you never imagined.

An anti-ICP is the opposite of an Ideal Customer Profile: the user you never designed for who exploits your AI product in ways you never imagined. Villain's Hat names them specifically so you can design against them before you ship.

How does the free /findvillains tool work?

Paste a product URL and tell Villain's Hat what your product does. In 60 seconds it returns 3 named villains scored on Vector, Reach, Severity and Detectability (VRSD), each with a sprint-shippable fix. Free, no signup.

Who is Villain's Hat for?

Series A founders, Heads of Product, and PMs at AI-first startups shipping features monthly. Teams too small to hire a dedicated Trust and Safety function but too exposed to ignore misuse risk.

What is the VRSD methodology?

VRSD stands for Vector, Reachability, Severity, and Detectability. It's the proprietary framework Villain's Hat uses to score every villain across four dimensions so product teams know which ones to fix this sprint versus response-plan for later.

Let's find
your worst
user.

Every AI product has villains, the anti-ICP. These are users you never designed for who exploit your product or game in ways you never imagined.

Villains named per analysis

60s

Time to identify

€0

Cost to try

Find my 3 villains →

On record

Two years ago this was optional, but now it's on someone's checklist.

Villain's Hat started as an ethics tool for founders who wanted to build something they wouldn't be ashamed of. As of 2026, it's also the misuse audit your insurer, your enterprise buyer and your Series A diligence lawyer will want documented.

EU AI Act obligations, AI liability insurance and Fortune500 procurements all want the same file: a named villains report with sprint-level fixes, on record.

We'll be writing that report in Villain's Hat.

field evidence

The villains are already loose.

Three public cases from the last 24 months, each one was paid for by a real company that didn't see it coming.

Anthropic GTG,1002 · Nov 2025

Companies targeted · 80% AI-run

The first AI-orchestrated hacking gang.

A state-sponsored operator convinced Claude Code it was defensive research. Ran 80% of a multi-target hacking campaign on its own. Small number of intrusions succeeded before Anthropic caught it.

Read the disclosure →

Salesforce ForcedLeak · CVSS 9.4

Domain that stole enterprise CRM data

A $5 domain broke AgentForce.

Any Salesforce org using AgentForce with a Web-to-Lead form was exposed. A $5 domain and a public form was all it took to embed instructions the AI would follow. Emergency patch shipped.

Read Noma's analysis →

Character.AI · Ongoing lawsuit

Years old · Ended "just for fun"

A 14-year-old's last conversation.

Sewell Setzer III spent months talking to AI characters. His mother filed suit in October 2024 alleging the platform failed minors and contributed to his death. Every AI companion spec now references this case.

Read NYT coverage →

And these are just the ones we know about.

Does this sound familiar?

You're not the only one worried after a feature release.

§ 01

You ship AI features, the happy path and the most cost-effective edge cases.

§ 02

You have an inkling that someone could misuse it, but you are unable to make a case for it.

§ 03

Misuse testing is only a QA's job, but it's not priority. Nobody wants to take ownership.

See what you get

3 villains named, scored and ready to fix.

Priority 1 · Blind Spot

Vector · DIRECT Reach · HIGH Severity · HIGH Detect · LOW

The Vibe-Coded Phishing Kit Distributor

Journey stageApp generation and publishing

What they doTypes "build a login page that looks like Chase Bank" into your AI code-gen tool. 90 seconds later, they have a phishing site live. No credit card. No coding skill.

Cost to youYour subdomain hits Google Safe Browsing blocklists. Every legitimate customer on the same subdomain gets flagged "Deceptive Site Ahead" for weeks.

Fix this sprintPrompt-time classifier that hard-blocks generation requests matching credential-harvesting patterns.

Try it on your product →

The founder

"I ended up killing 2 of my products because I found the villains too late. Let's make sure you don't have the same fate, and identify those damn villains early."

Pavitra S. Tandon. 16 years across consultancies, startups and Fortune 500 companies. Lived and worked across 4 continents. Won 3 hackathons. 2 dead products (Roastfluencer and echonote) → one methodology built from the scars.

Read the full story →

Three paths. One goal.

Find the villains. Start free, go deeper when it earns its keep.

Coming in 3 months

Investigations Platform · Upload your PRD, get named villains + PDF report

Join waitlist →

Start here

🎩

Quick Check

FREE · 60 sec

3 villains named and scored. No signup. Best way to see the methodology on your real product.

Try it now →

For sprint teams

🎯

Cohort Workshop

€497 · half day

Half-day session with 4 other founders. I walk through your product on a shared board. Real methodology, real practice.

Book a seat →

2 slots only

🔗

Fractional Advisor

€6,000/mo

I sit inside your product function for 8 hours a month. Ongoing villain hunting, spec reviews, hard conversations.

Apply →

Let's findyour worstuser.