AI agents fail. Learn why.

Every Incident, Decoded

Searchable database of 1,000+ real failures from Claude, GPT, o1, and other coding agents. Know the edge cases before they ship to production.

Search Incidents

When AI Coding Agents Fail

Claude, GPT, o1 - they ship broken code. Not often. But predictably. Hallucinated imports. Off-by-one loops. SQL injection in parameterized queries. Type confusion. Async traps.

Every failure is a pattern. Every pattern is avoidable. But only if you know it first.

This database collects real incidents from production, captured from Cursor, VS Code AI, agentic workflows, and code generation at scale. See what breaks. See why. See how teams defended themselves.

Hallucinated Import. Claude generated `import UseMerge from '@/hooks/merge'` but the module never existed. Code compiled. Failed at runtime in production.

SQL Placeholder Order. GPT-4 wrote parameterized SQL with placeholders in wrong order: `WHERE user_id = $2 AND org_id = $1` when args were `[org_id, user_id]`. Silent data leak for weeks.

Off-by-One in Normalization. Claude's vector normalization skipped the final chunk. ~0.3% of embeddings were wrong. Found only after 6 months in production search.

Promise Without Catch. o1 generated async chain with no error handler. One API failure cascaded to 10,000 orphaned records. No logs. No alerts.

Failure Categories

Hallucinations

Imports, functions, APIs that don't exist. Generated code compiles but fails at runtime.

Type Confusion

Strings sent where ints expected. Silent type coercion. Broken contracts at runtime.

Logic Gaps

Off-by-one errors. Empty list handling. Boundary conditions. Silent numeric failures.

Security Blind Spots

SQL injection in parameterized queries. Auth bypass. Input validation gaps. Race conditions.

Performance Cliffs

O(n²) loops. Unbounded recursion. Memory leaks. Cache invalidation that doesn't.

Async Traps

Missing awaits. Promise chains without error handlers. Race conditions in concurrent code.

Search by Pattern, Agent, Language

Every incident is indexed by: agent type (Claude, GPT, o1), framework (React, FastAPI, Rails), language (JavaScript, Python, Go), and failure pattern (hallucination, type error, security).

Filter by severity to see the most costly failures first. Or browse by agent to know exactly what breaks in Claude Code, GPT-4, and others.

Database live at ihavebeenclawed.com
Search is free. Full-text incident context, code samples, and defense patterns. Trending analysis across 1,000+ incidents shows you what's breaking right now in the ecosystem.

Want I Have Been Clawed built out, or handed over?

This is not a purchase and there is no card field. It puts your address, this product, and whatever you write below in front of a person, and you get a written answer about what finishing it, or handing it over for you to run yourself, would actually take.

A person reads these. Expect a reply in days, not minutes.
Your address is used to answer you. No sequence, no list, no sharing.
Ask to be removed and you are removed.