Agent Failures Index · Monthly report · As of
AI agent failures in June 2026
How many AI agent failures were there in June 2026?
The Agent Failures Index records 7 AI agent failures that happened in June 2026. The most common types were rogue action (2), prompt injection (2) and data exposure (1). None were case studies of organisations hurt by their own agents, and 6 were critical or high severity. The most serious: LiteLLM proxy authentication bypass via Host header injection.
By failure type
| Type | June 2026 |
|---|---|
| Rogue action | 2 |
| Prompt injection | 2 |
| Data exposure | 1 |
| Agent-enabled attack | 1 |
| Reliability | 1 |
By class
| Class | Incidents |
|---|---|
| Security exploit | 6 |
| Tool defect | 1 |
Every incident in June 2026
Counted by the date each incident happened, from the Agent Failures Index. Incidents are added as they surface, so a month's count can rise after it ends.