Why SHA-256 Isn't Enough for AI Agent Governance (And How We Fixed It)

Outline & Key Points:

The Problem: Explain why self-signed SHA-256 digests fail as authentication (anyone with write access can recompute). Use the audit's finding: "integrity ≠ authentication".

The Threat Model: Briefly describe attack vectors (bundle tampering, rollback, unauthorized issuer).

The Solution: Walk through Ed25519 signing + trust store verification. Show code snippets of signing.py and runtime verification.

Beyond Signing: Mention ActionRequest model and unified integrations as complementary hardening measures.

Why SHA-256 Isn't Enough for AI Agent Governance (And How We Fixed It)

Other newsrooms on this story

Related reading

I Discovered AI Agents Can't Self-Verify. The Real Problem Is Much Bigger.

# How we built a tamper-evident WORM audit log for AI agents using SHA-256 hash…

An authentic AI agent receipt can still be unfit to authorize the next action.

Two signatures are better than one — bilateral provenance for AI agents

I built a cryptographic passport for AI agents — here's how it works

Cool AI Projects That Failed: The File Integrity Gap