PricingDemo
Log InGet API Key
AI Agent Security — Cryptographic Attestation

Prove What Your
Agent Did

AI agents make decisions, access data, and invoke services autonomously. HATS produces cryptographic proof of every action — independently verifiable, post-quantum signed.

Schedule Demo HATS Standard

The Problem

Autonomous AI agents operate without human oversight at execution time. When something goes wrong, current verification approaches fail.

📝

Logs Are Self-Reporting

Agents write their own logs. A compromised or misconfigured agent can omit, reorder, or fabricate log entries. Self-reported activity is not evidence.

🔍

No Proof of Data Access

Did the agent read PII? Access medical records? Query a restricted database? Without cryptographic attestation, there is no verifiable answer — only the agent's claim.

🔗

Multi-Agent Chains Are Opaque

Agent A calls Agent B calls Agent C. Who touched what data? Who authorized the delegation? Current architectures have no provenance chain — just nested API calls.

📄

Insurance Won't Cover Unverifiable AI

Cyber insurers are excluding AI agent actions from coverage. Without independently verifiable proof of agent behavior, liability shifts entirely to deployers.

How Agent Attestation Works

Every agent action generates a 74-byte cryptographic commitment. The full chain is independently verifiable without access to the agent system.

1. Agent receives task + delegated authority from orchestrator
   
2. Each data access generates H33-74 attestation (74 bytes, PQ-signed)
   
3. Each computation attested — FHE if sensitive data (encrypted in-use)
   
4. Delegation to next agent: previous attestation verified first
   
5. Full provenance chain: independently verifiable by any third party

No system access required for verification. Auditors, regulators, and insurers verify agent behavior using only the 74-byte H33-74 commitments and the public verification endpoint. No VPN, no agent credentials, no infrastructure access.

Multi-Agent Provenance Chains

When agents delegate to other agents, each link in the chain is cryptographically bound. The chain breaks precisely at the point of compromise.

Agent A
Retrieves patient data
Agent B
Analyzes risk factors
Agent C
Generates recommendation
Each agent verifies the previous agent's attestation before proceeding. No valid attestation = no delegation.
1. Agent A attests: data accessed, authority source, output hash
2. Agent B receives task + Agent A's attestation
    Verifies A's attestation before executing
    Generates its own attestation (includes ref to A's commitment)
3. Agent C receives task + Agent B's attestation
    Verifies B's attestation (which transitively verifies A)
    Generates its own attestation (full chain committed)

Chain breaks at point of compromise. If Agent B's attestation is invalid — wrong authority, tampered data, expired delegation — Agent C refuses to proceed. The failure is localized and attributable. No silent failures, no ambiguous blame.

What Gets Attested

Every attestable event produces a 74-byte H33-74 commitment: 32 bytes on-chain, 42 bytes in Cachee. Three post-quantum signature families.

Event TypeWhat Is RecordedH33-74 Commitment
Data Access Which data sources the agent read, field-level scope, timestamp 74 bytes — PQ-signed
Computation Type Operation performed (inference, aggregation, comparison, transformation) 74 bytes — PQ-signed
Delegation Authority Who authorized the agent, scope of delegation, expiry 74 bytes — PQ-signed
Policy Compliance Which policies were evaluated, pass/fail status, policy version hash 74 bytes — PQ-signed
Output Hash SHA3-256 hash of agent output — proves output was not modified after attestation 74 bytes — PQ-signed
Timestamp Attestation creation time, chain position, ordering proof 74 bytes — PQ-signed

74 bytes per event. Three post-quantum signature families (ML-DSA + FALCON + SLH-DSA) compressed into 74 bytes via the H33-74 substrate. Breaks only if MLWE lattices, NTRU lattices, and stateless hash functions are all simultaneously broken — three independent mathematical bets.

HATS Integration

HATS is a publicly available technical conformance standard for continuous AI trustworthiness; certification under HATS provides independently verifiable evidence that a system satisfies the standard's defined controls.

Continuous Monitoring

HATS does not audit after the fact. It continuously monitors agent behavior in real time, generating attestations as actions occur — not reconstructing them from logs later.

Cryptographic Proof, Not Logs

Every HATS control produces a cryptographic commitment. Compliance is not a checkbox — it is a verifiable chain of post-quantum signed attestations that any third party can independently validate.

Agent-Native Controls

HATS defines controls specifically for autonomous AI agents: delegation authority verification, data access scope enforcement, multi-agent chain integrity, and output attestation.

Learn About the HATS Standard See the HATS Demo

Industry Applications

AI agents are deploying across regulated industries. Each has specific compliance requirements that agent attestation addresses directly.

Healthcare

HIPAA Agent Compliance

Prove that an AI agent accessing patient records stayed within authorized scope. Attest every PHI access event. Provide verifiable evidence for HIPAA audits without exposing patient data.

Finance

Fiduciary Proof

AI agents making investment decisions, executing trades, or assessing credit need verifiable proof that they operated within fiduciary bounds. Attestation creates the audit trail regulators require.

Legal

Chain of Custody

AI agents reviewing documents, conducting discovery, or generating summaries create chain-of-custody concerns. Attestation proves which documents were accessed and what operations were performed.

Insurance

Verifiable AI for Underwriting

Cyber insurers need to assess whether AI agents operated within policy terms. Agent attestation provides the independently verifiable evidence required for claims processing and coverage decisions.

Verify It Yourself