Blog
Notes on securing AI agents with deterministic policy: signed manifests, enforcement, and tamper-evident audit.
3 min read
A model can be persuaded. A signed policy cannot.
Prompt rules are not a security boundary for AI agents. Four ideas on moving the decision outside the model — and a rule you can write today.
securitypolicyprompt-injection
2 min read
What a hash-chained audit log actually proves
Tamper-evident is not tamper-proof. Four ideas on what a chained log guarantees, what it does not, and how to check your own.
auditintegrityspec