Ledger / Aug 2026
Attacker prompts an exposed agent app into revealing METR's API key
Strong
METR, the victim, described the incident; it reaches us through AI Threat Watch.
- Model
- Not namedAI agents (model not named)
- Agent harness
- Nonechat or tool use only
- Attributed country
- Not attributedthe source names no country
- Sectors hit
- ai platformsMETR (AI evaluation nonprofit), public-models API account
A fail-open bug exposed a researcher's agent app to the internet; an attacker prompted an agent to reveal METR's API key and used it for three weeks.
“A fail-open bug in a researcher's AI-written agent app exposed it to the internet; an attacker prompted an agent to reveal METR's public-models API key and used it for three weeks.”AI Threat Watch
Sources
- Attacker prompts an exposed agent app into revealing METR's API keyReported via AI Threat WatchAI Threat Watch · Aug 2026
- Date
- Aug 2026
- Actor
- Unknown external attacker (unknown), as attributed by AI Threat Watch
- Targets
- METR (AI evaluation nonprofit), public-models API account
- What the AI did
- Agent hijack, LLMjacking
- Model as named
- AI agents (model not named)
- Model basis
- The victim says so
- Autonomy
- Assisted a human
- Outcome
- The key was used for three weeks; METR puts the free credits at about $600,000 and says no sensitive data was accessed.