Posts
Findings from the tagged incident data, newest first. Every chart in a post is the live chart from stats.
- L6Identity / Run
Where agent breaches actually start: 385 incidents, eight layers, one surprise
I tagged every disclosed AI agent security incident I have cataloged since 2024 against an eight-layer provenance model. The leading cause is not prompt injection and it is not stolen keys. It is the absence of an identity boundary around the agent run.