Ledger / 8 Sep 2026
Exposed 'Recon' framework used AI agents to harvest 23,800 secrets
Strong
Google found agent config files and a dashboard on an exposed C2 server.
Agent config files such as AGENTS.md and agentic_vuln_research.md showed AI agents researching vulnerabilities and exploiting servers with minimal human input.
“By leveraging autonomous AI agents to research vulnerabilities, scan server-side infrastructure, and execute targeted exploits”Google Threat Intelligence Group
Sources
- GTIG AI Threat Tracker: From Prompting to AutonomyGoogle Threat Intelligence Group · 8 Sep 2026 · primary source
- Disclosed
- 8 Sep 2026
- Activity
- 2026
- Actor
- Unattributed (criminal), as attributed by Google Threat Intelligence Group
- Targets
- Server-side infrastructure; cloud and AI service keys
- What the AI did
- Agentic intrusion, Vuln research, Ops automation
- Model
- unspecified LLM agents (OpenClaw-style agent files)
- Model family
- Unspecified
- Model basis
- stated by vendor
- Autonomy
- autonomous
- Outcome
- Google disabled assets; over 23,800 secrets managed in the dashboard.