The Cryptographic Standard for AI Agent Accountability

The trust layer
for AI agents.

Every AI agent action needs a cryptographic record proving who authorized it, what the constraints were, and whether they were honored. Existing protocols were built for humans — not autonomous agents. IPP was built for this.

Protocol Stack
Your AI Agent LangChain · AutoGen · CrewAI · Custom
Application
Intent Provenance Protocol Intent Tokens · Genesis Seal · Narrowing Invariant
IPP ← Here
Cryptographic Identity Ed25519 · DIDs · TLS
Identity
Transport / Infrastructure HTTP · gRPC · Cloud APIs
Transport
Published & Filed
IETF Internet-Draft  draft-haberkamp-ipp-00
Specification  ipp.khsovereign.com/spec/v0.1
Open Source SDK  github.com/khsovereign
License  CC BY 4.0 — Open Standard
The Problem

OAuth and SAML were
built for humans.
Agents are not.

Every enterprise security framework built in the last thirty years assumes a human is the actor at the center of every consequential action. AI agents have invalidated that assumption entirely.

82:1 Machine to human identities in the average enterprise
80% Of IT professionals who have seen AI agents act outside authorization
EU AI Act · Article 14 Organizations must prove every AI action was authorized at time of execution — not just when credentials were issued.
Aug 2026
01

No human in the loop at authorization time

OAuth requires a human to click a consent screen. AI agents running autonomously for days cannot do this. Developers work around it with broad pre-granted scopes and shared service accounts — creating exactly the exposure OAuth was designed to prevent.

02

Static scopes cannot encode bounded intent

When a CFO says "optimize cash positions and move idle balances over $10M into short-term treasuries," that instruction contains a goal, a constraint, and a boundary. OAuth's scopes — "read email," "write calendar" — were never designed to carry machine-enforceable human intent.

03

Delegation chains have no enforcement

When one agent spawns a sub-agent, there is no protocol-level mechanism ensuring the derived agent cannot exceed the original human's intent. Each delegation step is a potential scope expansion with no cryptographic check.

04

No provenance — only permissions

OAuth logs tell you what an agent was allowed to do. They cannot tell you why it did it, who authorized it, what the constraints were, or who is legally accountable. That is a provenance gap. And it is the one regulators are about to close.

How It Works

Four properties. One protocol.

IPP introduces four foundational properties enforced through Ed25519 digital signatures, Decentralized Identifiers, and the Narrowing Invariant — a novel enforcement mechanism that makes scope expansion cryptographically detectable.

Lineage

Every action taken by every agent is traceable through an unbroken chain of cryptographic signatures to a human Principal — a person with a name, a legal jurisdiction, and accountability. Not a service account. A person.

Boundedness

Every Intent Token carries explicit, machine-readable constraints on authorized scope. Those constraints travel through every delegation level and cannot be expanded by any intermediate agent. The Narrowing Invariant enforces this cryptographically.

Non-Repudiation

Every token is cryptographically signed. The record is verifiable by any third party — auditors, regulators, counterparties — without requiring communication with the original issuer. Legally defensible by design.

Interoperability

Compliant implementations work regardless of AI framework, cloud environment, or programming language. Any conformant implementation can verify any conformant token. Framework-agnostic. Cloud-agnostic. Open standard.

Reference SDK

Three lines.
Any agent.

The IPP Python SDK adds cryptographic governance to any existing LangChain, AutoGen, or CrewAI agent without modifying its logic. Drop it in. Every action is now governed, audited, and provably authorized.

$ pip install ipp-sdk
View on GitHub →
agent_integration.py
# 1. Create a root Intent Token — once per task token = IntentToken.create( principal=Principal.from_env(), intent="Optimize cash positions", domain="financial.treasury", prohibited_actions=["equity_purchase"], expires_in="8h" ) # 2. Verify the Genesis Seal token.verify_genesis_seal() # 3. Record an action in the provenance chain token.append_provenance( action_type="financial.treasury.transfer", action_summary="Wire $12M to Treasury #7892", outcome="success" )
Protocol Comparison

What existing protocols
cannot do.

Requirement OAuth / SAML Intent Provenance Protocol
Human authenticationStrongComplementary layer
Bounded intent — machine-enforceableNot supportedCore primitive
Delegation chain — cryptographicNot supportedNarrowing Invariant
Scope narrows — cannot expandNot enforcedProtocol-enforced
Action provenance — append-only auditNot supportedProvenance chain
Legal attribution — defensibleNot supportedNon-repudiation by design
Cross-org trust — no central authorityRequires federationDecentralized verification
Revocation — mid-chain propagationToken-level onlyFull ancestry revocation
Authorship — permanent cryptographic recordNot supportedGenesis Seal
Important Distinction

IPP is not an identity platform. It is the intent provenance standard that identity platforms implement.

Identity and access platforms govern who an agent is and what it can access. IPP governs why it acted, under whose bounded human intent, and provides the cryptographic proof that the action stayed within those bounds. These are complementary layers — IPP sits beneath identity platforms and above the cryptographic infrastructure they depend on.

The Founder

Built by a
practitioner.

A
Amanda Haberkamp
Founder & CEO, KH Sovereign, Inc.

17 years in cybersecurity and spent my career watching enterprises struggle to answer basic accountability questions about their systems.

When AI agents arrived, the same gap appeared — but at a scale and speed that existing protocols cannot address. So I built the infrastructure layer to fix it.

My name is embedded mathematically in every token this protocol produces. It cannot be removed.

March 2026
KH Sovereign, Inc. founded
Intent Provenance Protocol v0.1 published as an open standard. IETF Internet-Draft draft-haberkamp-ipp-00 filed. Key Generation Ceremony completed — Genesis Seal cryptographically signed and committed to GitHub.
March 2026
Reference Python SDK published
pip install ipp-sdk. Three lines of code adds cryptographic governance to any LangChain or AutoGen agent. 38 passing tests. JavaScript SDK in development.
Now
In active conversations with enterprise security teams
Working with Fortune 500 security leaders to validate the protocol against real AI agent deployments. Targeting $4M seed round. Building the provenance standard before the accountability crisis — not after it.
August 2026
EU AI Act Article 14 enforcement begins
Organizations must prove every AI-driven action was authorized at time of execution. OAuth logs will not satisfy this requirement. The compliance window is closing.
Writing

From the blog

All posts →
KH Sovereign, Inc.

The accountability
gap is open now.

Every enterprise deploying AI agents today is accumulating compliance exposure they cannot yet address. The window to build the infrastructure is closing.