The book starts with the problem, AI agents are writing and executing code with minimal oversight. Databases deleted, drives wiped, dozens of CVEs across every major AI coding tool. Usage is up, trust is down.
Then it walks through one approach, Ed25519 signed permits, Secure Enclave integration, M of N team approval, capability gated runtimes, and an ECDH killswitch. It stirs ideas about what the future may look like but doesn't necessarily have to but stirs the idea of what the gates might be then.
Would appreciate feedback from anyone working on similar problems & your approach.
Book: https://book.se.tc page: https://se.tc Docker: docker pull humanatsetc/setc:book-edition