Agent Failures Index · Monthly report · As of
AI agent failures in August 2026
How many AI agent failures were there in August 2026?
The Agent Failures Index records 8 AI agent failures that happened in August 2026, against 16 in July 2026. The most common types were prompt injection (2), rogue action (2) and oversight gap (1). 3 were case studies of organisations hurt by their own agents, and 4 were critical or high severity. The most serious: Eclipse Theia AI Agent Mode file-write tools let a model-supplied path escape the workspace and hit remote code execution.
By failure type
| Type | August 2026 | July 2026 |
|---|---|---|
| Prompt injection | 2 | 6 |
| Agent-enabled attack | 1 | 4 |
| Rogue action | 2 | 2 |
| Oversight gap | 1 | 1 |
| Accuracy | 0 | 2 |
| Abandoned deployment | 1 | 0 |
| Reliability | 1 | 0 |
| Data exposure | 0 | 1 |
By class
| Class | Incidents |
|---|---|
| Security exploit | 4 |
| Production incident | 2 |
| Programme failure | 1 |
| Tool defect | 1 |
Every incident in August 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.