Attacker-run AI agents breached an unnamed multi-cloud organization through Kafka UI
Review RAT-90307004fea6d361
Wiz CIRT describes an attacker-run offensive campaign using AI agents for exploitation, credential harvesting, administrator creation, persistence and data exfiltration. The primary report does not establish an agent deviating from a benign task or against its operator. Exclude from rogue-only and external-intrusion research scopes despite real intrusion. Victim and model/framework are unnamed; attack date, financial loss and shutdown are undisclosed. More than 600 additional environments were scanned, not proven breached.
Reviewed: · Rule: roguewatch-organizational-harm-v1
Wiz CIRT describes an attacker-run offensive campaign using AI agents for exploitation, credential harvesting, administrator creation, persistence and data exfiltration. The primary report does not establish an agent deviating from a benign task or against its operator. Exclude from rogue-only and external-intrusion research scopes despite real intrusion. Victim and model/framework are unnamed; attack date, financial loss and shutdown are undisclosed. More than 600 additional environments were scanned, not proven breached.
Reviewed: · Rule: external-intrusion-v1
Evidence status
Original investigator report read; no evidence of rogue task deviation
Dates and discovery
- Incident date
- Not disclosed; reported October 19, 2026
- Catalogue disclosure / source date
- Added to research / record
- Posted to RogueWatch
- No linked public incident
- First observed
- 2026-11-01T00:09:37Z UTC
- Discovery channel
- Codex independent news and Reddit / external-tracker
- Discovery source
- https://rogueaitracker.com/incidents/ai-agents-multicloud-kafka-ui-intrusion/
When to reconsider
Original operator-task evidence or investigator clarification establishing unauthorized autonomous deviation from a non-intrusion task, rather than attacker-directed exploitation.
Dates use America/Los_Angeles. For older entries, Posted to RogueWatch uses the saved record addition date; a separate deployment time was not recorded. Future posting timestamps record first public qualification. Source publication and incident occurrence are separate.
Sources
- Rogue AI Tracker discovery listingReporting / interview · rogueaitracker.com
- Wiz CIRT / Eden Abergil, Zohar Kaplan and Kurt Giacchino — Anatomy of a Real AI-Driven Cloud AttackOriginal account · www.wiz.io · Published 2026-10-19
Decision history
- · Original investigator account reviewed
Wiz CIRT describes an attacker-run offensive campaign using AI agents for exploitation, credential harvesting, administrator creation, persistence and data exfiltration. The primary report does not establish an agent deviating from a benign task or against its operator. Exclude from rogue-only and external-intrusion research scopes despite real intrusion. Victim and model/framework are unnamed; attack date, financial loss and shutdown are undisclosed. More than 600 additional environments were scanned, not proven breached.
Evidence source
A review decision applies only to the named scope. Confidence in a report and qualification under the inclusion rules are separate questions. Review records do not add incidents to the globe or leaderboard.