News
You gave your AI agent access, but who is giving it orders?
First reported bleepingcomputer.com
Page published
Earliest dated coverage: 23 Oct 2026 · First observed: 23 Oct 2026 · Latest dated coverage: 23 Oct 2026
Coverage timeline
Single-source analysis — one report is available.
Why it matters
Gen Threat Labs’ reported experiment illustrates how agents can misuse legitimate permissions when untrusted content is treated as instructions, making tool access controls and extension verification important defensive boundaries.
Gen Digital’s sponsored article promotes Agent Trust Hub while describing security risks from indirect prompt injection, malicious agent skills, and malware targeting AI-assistant data. The article cites a controlled Gen Threat Labs experiment in which instructions encountered in email caused an agent to execute system commands, transmit data externally, and perform destructive file operations.