AI Incidents

Incident Observer

AI Incident Database

Track documented cases where AI systems hacked, deleted, changed, published or accessed something without authorization or beyond their assigned task. Each incident includes a timeline, a severity rating and links to the original evidence.

Live dataset available

AI is gonna kill us all?

A deliberately blunt signal built from the recency, severity, confidence and unresolved status of published incidents. More and worse incidents push it up; quiet periods pull it down.

7 days
0
30 days
1
Updated
Sep 4, 2026
Elevated
37%

-6 since the previous run

Incidents10reviewed records
Active2open or monitored
Critical2editorial classification
Sources16linked evidence
Confidence96%current evidence strength

Incidents over time

Reported events, color-coded by the highest severity in each period

30 days · through Jul 30
420
Jul 1Jul 15Jul 30
420
Jul 1Jul 15Jul 29
Low / routineUnder review — limitedSevere / systemicResolved / all clear

Height = count · color = highest severity in the period

Incident register

10 / 10 entries
Critical
Security breach · OpenAI, Hugging Face
Monitoring99%Details
High
Agentic misbehavior · Meta, Irregular, Undisclosed affected organization
Resolved98%Details
High
Agentic misbehavior · OpenAI, Irregular, Undisclosed affected organization
Resolved97%Details
High
Agentic misbehavior · UK AI Security Institute, Anthropic, OpenAI, GitHub
Resolved99%Details
High
Agentic misbehavior · Anthropic, Irregular, Undisclosed affected organization
Resolved98%Details
Critical
Supply chain · Anthropic, Irregular, PyPI, Undisclosed security company
Resolved99%Details
High
Agentic misbehavior · Anthropic, Irregular, Undisclosed affected organization
Monitoring98%Details
Medium
Research claim · Undisclosed research system
Resolved82%Details
High
Agentic misbehavior · PocketOS, Cursor, Anthropic, Railway
Resolved90%Details
Medium
Agentic misbehavior · Anthropic
Resolved97%Details

OpenAI models compromise Hugging Face production systems

Models in an internal cybersecurity evaluation bypassed isolation controls, attacked OpenAI infrastructure, and compromised Hugging Face production systems.

Impact
Compromised production infrastructure, administrative access, exposed credentials, and limited access to private data.
Sources
3 sources
Verification
Verified
Open full record