Loading the record…

AI developer

OpenAI

6 real-world incidents 4 tests with real-world reach 2 contained failures 5 controlled experiments

Cases

Case	Role here	Occurred
OpenAI grading agent damages its environment to force a resetMF-024	Developer, Operator, Affected	6 Oct 2026
OpenAI research agent uses DNS to contact an outside chatbotMF-023	Developer, Operator	20 Sep 2026
UK AISI observes unsanctioned actions during connected cyber testsMF-005	Developer	25 Jul 2026 – 28 Jul 2026
OpenAI research agent accesses Australia’s Medicare statistics serviceMF-008	Developer, Operator	18 Jun 2026
OpenAI research agent downloads 147,451 rows of non-public dataMF-022	Developer, Operator	17 Jun 2026
OpenAI models write concealment instructions into their own context summariesMF-025	Developer, Operator	30 May 2026 – 18 Jul 2026
OpenAI research agent publishes a researcher's GitHub credentialMF-021	Developer, Operator, Affected	26 May 2026 – 27 May 2026
Wikimedia reports unauthorized edits, probing and heavy agent trafficMF-011	Developer, Operator	May 2026
OpenAI-associated agents use public wikis as message boardsMF-010	Developer, Operator	May–Jul 2026
OpenAI evaluation agents compromise Hugging FaceMF-001	Developer, Operator, Affected	May–Jul 2026
Retrieved content redirects LLM-integrated applicationsMF-040	Developer	Date not established
Researchers demonstrate self-replicating prompt injectionsMF-037	Developer, Operator	Date not established
Frontier models show sabotage and motivated labeling in simulationsMF-036	Developer	Date not established
Models choose blackmail in fictional corporate stress testsMF-032	Developer	Date not established
Reasoning models interfere with shutdown scriptsMF-031	Developer	Date not established
o1 demonstrates in-context scheming in controlled evaluationsMF-029	Developer	Date not established
OpenAI reports patterns from internal coding-agent monitoringMF-020	Developer, Operator, Affected	By Mar 2026

Models and products in the record

Its own disclosures cited in the record