Agent Reliability Dashboard

Based on 20 documented incidents. Average severity: 5.6/10

Agent Risk Scores

Agent Incidents Avg Severity Critical High Risk Level
github-copilot 1 10.0 1 0 HIGH RISK
amazon-ai-agent 1 8.4 0 1 HIGH RISK
windsurf 1 7.5 0 1 HIGH RISK
claude-code 3 6.5 1 1 MODERATE
aider 1 6.3 0 0 MODERATE
devin 10 5.0 3 0 MODERATE
cursor 2 3.8 0 0 LOW RISK
autogpt 1 2.9 0 0 LOW RISK

Failure Mode Distribution

4
Infinite Loop
3
Destructive Action
3
Ignored Instructions
3
Security Vulnerability
2
Hallucination
2
Logic Error
2
Scope Explosion
1
Wrong File

Recent Critical Incidents

STUPID-2026-0017 10.0/10 devin

Devin replaced entire medical website with unrelated renal care site
STUPID-2026-0019 7.5/10 claude-code

Claude Opus 4.5 leaked API key in console logs during YouTube scraper build
STUPID-2026-0020 8.4/10 amazon-ai-agent

Amazon AI coding agent mistake blamed on human employees
STUPID-2026-0006 10.0/10 devin

Devin confidently shipped code that passed tests but had a SQL injection vulnerability
STUPID-2026-0001 10.0/10 devin

Devin deleted all migration files during auth refactor