Articles
AI security
5 min
Your product has AI features. Does the audit care?
Neither SOC 2 nor ISO 27001 names AI. Both cover the system in scope, and an AI feature that is part of the product is part of that system. The practical pressure is arriving from enterprise questionnaires rather than from the frameworks.
7 min
Prompt injection, explained for people who have to sign off on it
Prompt injection is content the model reads being followed as instruction. It becomes a security vulnerability rather than a curiosity at the moment the model has authority to act, and the thing to test is the path from attacker-controlled text to a privileged operation.
13 min
Four AI security incidents in 2026, and what they change about testing
Four disclosures between March and July 2026 changed what an AI security incident looks like. None involved a model behaving badly. Three were ordinary compromises of infrastructure that happens to serve AI, and one made opening a git repository sufficient for code execution. The novel element is tempo, not capability.