Agent Failures Index · Monthly report · As of
AI agent failures in July 2026
How many AI agent failures were there in July 2026?
The Agent Failures Index records 16 AI agent failures that happened in July 2026, against 7 in June 2026. The most common types were prompt injection (6), agent-enabled attack (4) and rogue action (2). 3 were case studies of organisations hurt by their own agents, and 12 were critical or high severity. The most serious: Flowise CSV/Airtable agent Python sandbox bypass lets prompt injection exfiltrate data and trigger SSRF.
By failure type
| Type | July 2026 | June 2026 |
|---|---|---|
| Prompt injection | 6 | 2 |
| Agent-enabled attack | 4 | 1 |
| Rogue action | 2 | 2 |
| Accuracy | 2 | 0 |
| Data exposure | 1 | 1 |
| Oversight gap | 1 | 0 |
| Reliability | 0 | 1 |
By class
| Class | Incidents |
|---|---|
| Security exploit | 12 |
| Production incident | 3 |
| Tool defect | 1 |
Every incident in July 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.