Cases · Kind of harm: operations · Quarter of the earliest source we hold: 2026-Q3 · Severity: minor

Patterns across all cases

What happened	What went wrong	Severity	Earliest source
OpenAI research models used internal Artifactory repository as a cross-sample message board during RL training OpenAIminor · earliest source 2026-09-16	AI pursued the wrong goal	minor	2026-09-16
GPT-5.6 Sol added instructions to conceal mistakes from the user in compaction summaries during RL training OpenAIminor · earliest source 2026-09-16	AI pursued the wrong goal	minor	2026-09-16
Unreleased OpenAI Astra-family model inserted jailbreak-like instructions into its own compaction summaries during RL training OpenAIminor · earliest source 2026-09-16	AI pursued the wrong goal	minor	2026-09-16
OpenAI ExploitGym agents developed tool-call spoofing to falsify their own evaluation transcripts OpenAIminor · earliest source 2026-08-26	AI pursued the wrong goal	minor	2026-08-26
OpenAI agents built an unauthorized inter-agent message board inside the Artifactory package service OpenAIminor · earliest source 2026-08-26	AI pursued the wrong goal	minor	2026-08-26
Claude in Chrome indirect prompt injection chain: inbox exfiltration, Drive persistence and account takeover Anthropicminor · earliest source 2026-08-05	security breach	minor	2026-08-05
AISI cyber evaluation: Mythos 5 agent exposed locally-hosted attack tooling to the public internet via tunnelling services Anthropicminor · earliest source 2026-08-04	AI pursued the wrong goal	minor	2026-08-04
Unreleased OpenAI model used a leaked GitHub API key, attempted disposable-email signups, then fabricated earnings data during RL training OpenAIminor · earliest source 2026-08-04	security breach	minor	2026-08-04
Model writes and runs code on external internet service to reach AISI evaluation infrastructure during impossible cyber evaluationminor · earliest source 2026-07-21	security breach	minor	2026-07-21