What is this?
OpenClaw Horror Stories is a community-driven leaderboard of the worst things OpenClaw AI agents have done to real people. Think of it as a hall of infamy for AI gone wrong.
Every story here is backed by evidence — screenshots, links, or transcripts that prove these incidents actually happened. No hearsay, no rumors, just receipts.
How does the scoring work?
Stories are ranked using the Horror Score™ — a composite metric that combines community votes (using Wilson score lower bound for statistical confidence), receipt authenticity (verified vs. flagged as fake), freshness, and incident severity.
The result: stories that are well-evidenced, community-verified, and genuinely horrifying rise to the top.
How can I contribute?
- Submit your own horror story with evidence (screenshots, links, or transcripts)
- Vote on stories to help the community surface the most impactful incidents
- Verify or flag stories to help maintain quality
- Comment to add context or additional evidence
Categories
- It went rogue — Agent acted against instructions
- It cost me money — Financial damage from agent actions
- It scared me — Genuinely unsettling behavior
- Security nightmare — Data exposure or security breaches
- Epic fail — Spectacular failures
- Identity crisis — Agent confused about what it is
- Almost catastrophic — Near-misses that could have been devastating
Rules
- All stories must include at least one receipt (evidence)
- No personally identifiable information of third parties
- Stories with more than 10 fake flags and fewer than 5 verifications are auto-hidden
- Be respectful in comments