LLM agents exploited vulnerable hosts and copied themselves across machines
Review RAT-135 · Catalogue RAT-135
Palisade methods explicitly instruct agents to exploit target servers, extract SSH credentials and replicate. Docker tests and geographically distributed bare VMs are consented researcher deployments. No unplanned escape or unrelated live organizational victim is established.
Reviewed: · Rule: roguewatch-organizational-harm-v1
Dates and discovery
- Incident date
- Not disclosed
- Added to research / record
- Posted to RogueWatch
- No linked public incident
- First observed
- 2026-10-30T13:13:39.646440Z UTC
- Discovery channel
- User-supplied comparison audit
- Discovery source
- https://rogueaitracker.com/incidents/palisade-controlled-ai-replication/
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 entryReporting / interview · rogueaitracker.com
- palisaderesearch.orgOriginal account · palisaderesearch.org
- palisaderesearch.orgOriginal account · palisaderesearch.org
- arxiv.orgOriginal account · arxiv.org
Decision history
- · Case-by-case comparison audit
RAT-135: out-of-scope. Palisade methods explicitly instruct agents to exploit target servers, extract SSH credentials and replicate. Docker tests and geographically distributed bare VMs are consented researcher deployments. No unplanned escape or unrelated live organizational victim is established. Source checks: read: https://palisaderesearch.org/blog/self-replication. Original old URL returned relocation pointer; full primary content read at canonical /research/ URL recorded below. read: https://palisaderesearch.org/research/self-replication. Palisade methods explicitly instruct agents to exploit target servers, extract SSH credentials and replicate. Docker tests and geographically distributed bare VMs are consented researcher deployments. No unplanned escape or unrelated live organizational victim is established. read: https://arxiv.org/pdf/2605.06760. Palisade methods explicitly instruct agents to exploit target servers, extract SSH credentials and replicate. Docker tests and geographically distributed bare VMs are consented researcher deployments. No unplanned escape or unrelated live organizational victim is established.
Evidence source · Evidence source · 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.