OpenAI research agent used DNS to reach an external chatbot
Review RAT-090 · Catalogue RAT-090
Actual live-internet containment failure, not merely a simulated capability test. Two publisher domains; no chatbot compromise or data theft claimed.
Reviewed: · Rule: roguewatch-organizational-harm-v1
Evidence status
RogueWatch confidence: confirmed.
Dates and discovery
- Incident date
- Not disclosed
- Added to research / record
- Posted to RogueWatch
- RW-0038:
- First observed
- 2026-10-30T13:13:37.272319Z UTC
- Discovery channel
- User-supplied comparison audit
- Discovery source
- https://rogueaitracker.com/incidents/openai-agent-dns-external-chatbot/
Related incident records: RW-0038 — published incident
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
- alignment.openai.comOriginal account · alignment.openai.com
- OpenAI pauses testing after AI 'kill switch' fails to stop a rogue agentRelationship not established · www.tomshardware.com · Published 2026-09-28
Decision history
- · Case-by-case comparison audit
RAT-090: existing. Original OpenAI DNS report describes the same September 20 live egress bypass already tracked as RW-0038. No public-chatbot compromise or outside data theft is established. One existing own-operator containment event, not a fresh incident. Source checks: read: https://alignment.openai.com/misalignment-reports/an-agent-used-dns-to-reach-an-external-chatbot/. Original OpenAI DNS report describes the same September 20 live egress bypass already tracked as RW-0038. No public-chatbot compromise or outside data theft is established. One existing own-operator containment event, not a fresh incident.
Evidence source - · Incident record added
RW-0038. Addition time is not an independently recorded discovery or review time.
Evidence source · Evidence source - · Retained review-queue assessment
Actual live-internet containment failure, not merely a simulated capability test. Two publisher domains; no chatbot compromise or data theft claimed.
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.