Articles

A local knowledge graph of me — the real-life person, not a job title and not a role. Imagine a chat where the assistant already knows who you are. Read more

A legitimate code review hit a safety classifier. A generic model rewrote the same request and finished the job. No attacker. No special tools. Read more

Twenty risk categories across two lists. Intent capsules and purpose-bound tokens are now named mitigations. Not one category asks whether an authorized action should happen. Read more

OAuth, SPIFFE, and CAEP give agents an identity floor. Standards bodies must now define the shared governance layer for purpose, trust, trajectory, and should. Read more

Every fragment looked fine. The intent only existed when you put them back together. Read more

Deepfake candidates are beating one-time identity checks. Hiring needs zero trust: verify the same person at every stage, from first screen to day ninety. Read more

Approval prompts are alert fatigue with better branding. Humans belong above the loop — defining boundaries and resolving hard cases — while governance enforces at machine speed. Read more

The industry is securing managed agents. The harder problem is governing autonomy across humans, agents, and agentic AI. Read more