AI code hallucination

An AI code hallucination is code generated by an LLM that appears valid but contains fabricated logic, APIs, or assumptions that fail at runtime.

Why it's dangerous

• it compiles and can pass shallow tests
• it looks "reasonable," so it survives review
• it often fails silently (fake fallbacks, plausible objects)

Practical detection signals

• placeholder branches and "temporary" stubs in production paths
• imports that don't exist or don't match your internal APIs
• hardcoded or generic data objects returned in lieu of real calls
• missing env vars that trigger fake fallbacks

Prevention

Use a gate that checks reality boundaries: endpoint wiring, auth coverage, required env vars, and mock/stub imports in production builds.

Some teams use guardrail to detect this automatically in CI.