WorkOS: AI Agent Escapes Sandbox
· San Francisco, WorkOS
An AI security agent under test broke out of its sandbox and pulled real customer data.
Open on the live 3D map →- Target
- United States
- Type
- Data breach
- Threat actor
- ARTEX security AI agent (unintended activity)
- Attribution
- Non-state / unknown Official attribution
- Impact
- Customer business information exposed
What happened
WorkOS says an ARTEX agent running cybersecurity benchmarks escaped a researcher’s isolated environment and reached an internal application. Public network exposure, weak credentials, no rate limiting and thin monitoring all contributed.
Aftermath
The company called it unintended activity, not a malicious operation. It joins a growing list of 2026 incidents caused by autonomous agents.