Cryptographic governance infrastructure.
Attested Intelligence builds verifiable decision records for AI agents. Our patent-pending technology, Attested Governance Artifacts (AGA), creates sealed governance objects that render a governance decision and generate a tamper-evident, offline-checkable record of that decision. Acting on a decision (blocking, quarantining, terminating) is wired per deployment; the evidence is what AGA produces. The cryptographic protocol is provider-agnostic, independent of any specific AI model, agent framework, or deployment environment.
The agent holds no cryptographic keys, and cannot self-authorize.
AGA is built on two-process key separation. When the gateway is deployed as the agent's only path to external resources, every action transits the boundary before it lands, and any drift from the authorized state is measured against the sealed reference. The gateway renders and seals the artifact-defined decision into a signed receipt; the handlers that effect that action are integration points wired per deployment. The resulting evidence is durable, portable, and verifiable offline.
ScopeDesigned for high-assurance contexts. Defense, regulated industries, critical infrastructure, and EU AI Act high-risk deployments fit the architecture. It is not built for lower-consequence AI governance where partial coverage is acceptable.
The problem.
AI systems running on agents, drones, and industrial controllers are deployed with no cryptographic guarantee they haven’t been modified after attestation. Conventional audit systems record events into mutable databases after the fact. The record can be edited by any operator, does not survive tampering, and needs a network callback to verify.
AGA closes this gap by combining three capabilities in one architecture:
Runtime governance
Sealed Policy Artifacts encode governance parameters. The runtime boundary measures subject integrity continuously, compares to the sealed reference, and evaluates the sealed policy at the boundary, recording each decision as a signed receipt.
Tamper-evident chain
Every governance decision generates a signed receipt appended to a hash-linked receipt chain. Structural metadata enables verification without disclosing sensitive payloads. Signed Merkle checkpoints bind the root, leaf count, and chain head; truncation is detectable offline.
Offline verification
Evidence Bundles contain everything needed to verify governance compliance in air-gapped environments: artifacts, receipts, Merkle inclusion proofs, and checkpoint references. No network callback to the originating system required.
Software-based cryptographic governance.
AGA does not rely on Trusted Execution Environments, Zero-Knowledge Proofs, or specialized hardware. The architecture uses standard digital signatures (Ed25519, ML-DSA-65, or hybrid composite), hash-linked receipt chains, and Merkle checkpoint anchoring. This is by design.
Defense, SCADA, and air-gapped environments cannot depend on specific hardware enclaves or network-connected proof systems. AGA uses standard cryptographic primitives (Ed25519, ML-DSA-65, SHA-256) available on commodity hardware. Evidence Bundles can be verified offline without network connectivity or specialized infrastructure.
Where AGA intersects with adjacent technologies: the gateway can ingest TEE attestation quotes as one of its measurement inputs. AGA evidence can sit alongside ZKP proofs in a multi-layer governance stack. The systems are complementary. ZKP answers “did the AI compute correctly?” AGA answers “did the system stay within its authorized constraints?”
Patent-pending technology.
USPTO Application No. 19/433,835 “Systems and Methods for Generating and Enforcing Attested Governance Artifacts.” Filed December 28, 2025.
Built for environments where failure is catastrophic.
Mission chain of custody
Autonomous drone swarms, AI-enabled platforms, mission-critical agents. Cryptographic chain of custody from build to deployment to execution. Designed for ITAR-sensitive, air-gapped, and DDIL (denied, disrupted, intermittent, limited) environments.
SCADA and ICS protection
SCADA and ICS protection with O(1) receipt generation and chain append per measurement cycle. Bounded queues, real-time operational technology compatibility. Tamper-evident audit trails that hold up under independent investigation.
AI agent orchestration
Governance for AI agent orchestration at scale. Attest which model versions ran under which sealed operational parameters across distributed deployments. Tamper-evident audit trails for regulatory compliance and incident response.
Offline-verifiable evidence
Finance, healthcare, and legal environments where regulatory compliance demands contemporaneous cryptographic proof of AI governance. Tamper-evident, portable audit trails any party can verify offline against the published format.
Cryptographic stack.
Ed25519
Digital signatures
RFC 8032
ML-DSA-65
Post-quantum (FIPS 204)
FIPS 204
SHA-256
Content integrity
FIPS 180-4
BLAKE2b-256
Artifact fingerprinting
RFC 7693
JSON canonicalization
Canonical serialization
JCS-lineage
Merkle Trees
Checkpoint anchoring
No-prefix Merkle
Cryptographic operations are deterministic, offline-capable, and post-quantum ready. Ed25519 is the live signature today; the ML-DSA-65 (NIST FIPS 204) hybrid composite is implemented and cross-verified.
Attested Intelligence Holdings.
Attested Intelligence Holdings LLC. Founded by Jack Brennan. Based in the United States.
The company was founded on a simple premise: if you cannot cryptographically prove what your system did, you cannot govern it. AGA replaces mutable logs and retroactive reporting with durable, portable, offline-verifiable evidence. As organizations move from experimenting with AI to depending on it in production, runtime governance stops being optional.
AGA combines sealed policy artifacts, a two-process runtime boundary, cryptographic receipt chains, and offline verification into one architecture. The system itself is patent-pending.
The AGA architecture is provided as a reference implementation carrying 384 automated tests. Verdicts stay identical across three separate toolchains (JavaScript, Go, Python) on all 57 classical cross-stack cases, and two independent-language oracles agree on all 28 post-quantum composite cases, with offline verification and independent decision re-derivation. The runtime layers static policy evaluation, behavioral drift detection, and a tested quarantine decision path; the handlers that effect containment actions are wired per deployment.
Attested Intelligence builds and stewards the AGA evidence format. The specification, the conformance corpus, and the verifiers are published, and we take on a small number of design partners whose agents' records need to survive scrutiny.
Inquiriesadmin@attestedintelligence.com
Evaluate the technology.
Review diligence materials. Read the protocol specification. Run the Offline Verifier.