Why OAuth and SAML Cannot Govern AI Agents

These protocols were engineered for a world where a human being sits at the center of every consequential action. That world no longer exists. Here is precisely where they break — five structural failures that no patch can fix — and what the protocol layer beneath AI agents actually needs to look like.

Protocol coverage by requirement
RequirementOAuthIPP
Human auth
Bounded intent
Delegation chain
Provenance
Legal attribution
Mid-chain revoke
Authorship record
Coming Soon

The Narrowing Invariant — How IPP Prevents Scope Creep in Agent Delegation

Every time an agent spawns a sub-agent, scope can only narrow — never expand. The cryptographic mechanism that makes that guarantee enforceable without a central authority.

Amanda HaberkampRead →
Coming Soon

EU AI Act Article 14 and What It Actually Requires from Your Agent Infrastructure

August 2026 enforcement means organizations must prove AI actions were authorized at time of execution. This is a provenance requirement — not an authentication requirement.

Amanda HaberkampRead →
Coming Soon

What the Key Generation Ceremony Is — and Why We Did It

The Genesis Seal is only as trustworthy as the ceremony that produced it. A behind-the-scenes look at the air-gapped machine, the Ed25519 key pair, and the sealed envelopes.

Amanda HaberkampRead →
Coming Soon

The Five Questions No Enterprise Can Answer About Their AI Agents Today

Who authorized it? What were the constraints? Did it stay within bounds? Who is accountable? Can you prove it? Most organizations cannot answer any of these cryptographically.

Amanda HaberkampRead →

Get notified when new posts publish.

No newsletter cadence. Only when something worth reading is ready.