Wiz presents three investigated intrusions involving an exposed Flowise service, a Java operations gateway and an MCP-facing GraphQL backend. It proposes seven behavioral fingerprints for agentic activity, combining tool adaptation, natural-language artifacts and execution patterns, while describing humans as setting objectives and guardrails.
Why it matters
It gives defenders case-based detection hypotheses while exposing the uncertainty in identifying agentic activity from operational traces.
Key facts
As stated in the sources, with where to find them.
- The cases describe reuse of an exposed service to target other installations, cloud-environment probing, and secret theft with persistence across workloads. These are Wiz investigative accounts, not independent victim statements.Agentic Attacks Investigated by Wiz; Case Studies
- Wiz combines sustained variable cadence, reasoning artifacts and adaptive behavior as indicators. It does not provide a validated detector error rate or establish a model identity from cadence alone.AI-agent fingerprinting criteria
- The article separately recaps previously published incidents. Its Hermes detection illustrations are labeled attack simulations and should not be counted as additional observed intrusions.From ARTEX to Hermes; simulation captions
Findings that cite this record
Key questions this bears on
- How are attackers using AI agents in real operations?To run parts of intrusions at scale: vendors report a state-aligned actor and criminal groups using open-source agent frameworks and commercial models.
Sources
Related records
Oct 21, 2026
Oct 29, 2026
Oct 26, 2026
Oct 21, 2026
Oct 7, 2026
Oct 1, 2026