Post-Quantum AI Compliance Infrastructure
for a regulated world.

AI systems make decisions.
Nobody can prove what they decided.

The EU AI Act Article 12 requires high-risk AI systems to maintain automatic logging of autonomous decisions — tamper-evident, retained for at least six months, available to regulators on demand. High-risk obligations apply from 2 December 2027, as deferred by the Digital Omnibus (adopted June 2026).

Existing approaches fail in one of three ways: application-level logs are mutable and unverifiable; third-party audit tools bolt on after the fact and miss real-time decisions; blockchain solutions introduce unacceptable latency for production AI pipelines.

The penalty for non-compliance is up to 3% of global annual turnover — for a $10B institution, that is $300M per year.

One line of code.
Every decision attested.

Rubric embeds at the framework level — not the application level. A single call at startup instruments every LLM call, agent decision, and tool use automatically. No per-decision code. No codebase audit. No latency impact.

Each attestation receives a unique ID, is cryptographically signed with ML-DSA-65 (NIST FIPS 204 post-quantum signature standard), aggregated into a Merkle forest, and anchored to Hedera Consensus Service — a public, immutable, verifiable ledger with Byzantine fault-tolerant consensus.

Built for production scale.
Not a proof of concept.

Rubric uses a three-tier aggregation architecture that batches many decisions into a single on-chain anchor — production-validated at up to 1,366 decisions per HCS anchor.

Post-Quantum Cryptography

All signatures use ML-DSA-65 (CRYSTALS-Dilithium), standardized by NIST in FIPS 204 (August 2024). All signing paths use @noble/post-quantum, an independently audited pure TypeScript implementation; signing completes in ~17.5ms server-side, fully off the customer's request path. Attestations signed today remain verifiable against quantum adversaries.

Payload Privacy

On the Enterprise tiered pipeline, decision payloads are encrypted with AES-256-GCM before storage. The customer holds the encryption key — Rubric stores only the ciphertext and a cryptographic commitment. Rubric cannot read your AI decisions. Regulators receive the key from the customer directly for audit purposes.

Zero-Knowledge Proofs

Encryption hides data; zero-knowledge proofs answer questions about it without unhiding it. Rubric operates both, in production, on a dedicated prover node (Noir/Barretenberg, BN254, Poseidon2 commitments).

Inclusion proofs. Every tiered attestation resolves to a hybrid proof: a zero-knowledge proof that its batch root belongs to the Poseidon2 forest whose root is anchored on-ledger, combined with a classical SHA3 Merkle proof of the attestation's membership in that batch. We state the hybrid structure plainly because evidentiary infrastructure should be exactly as strong as it claims — no more, no less.

Policy proofs. For schema-v2 records (July 2026 forward), each attestation carries a Poseidon2 payload commitment binding its source, confidence, and risk level. From it, Rubric generates zero-knowledge proofs of policy compliance: this decision came from the claimed system, with confidence at or above X and risk at or below Y — while the decision data, exact confidence, and exact risk remain undisclosed. The construction cannot prove a false statement: a proof request against bounds the record does not satisfy fails at witness generation, by design. Proof roots are anchored to Hedera mainnet in every tier-2 anchor; verification is a public endpoint requiring no API key and no trust in Rubric.

Aggregate compliance proofs. The same machinery extends to populations: a proof that at least K of N distinct records under an anchored root satisfied policy P, with a circuit-level uniqueness guard making double-counting mathematically impossible. This yields compliance reports — “99.4% of Q3 decisions met bounds, proven, with no individual decision disclosed” — chained to named public-ledger sequence numbers. Generated by Rubric on request.

Where the frontier is. Rubric proves properties of post-quantum-signed records in zero knowledge; it does not re-verify ML-DSA-65 lattice signatures inside circuits. In-circuit verification of a single FIPS 204 signature costs on the order of millions of constraints (SHAKE and emulated modular arithmetic dominate); our commitment scheme delivers the equivalent evidentiary binding in a handful of Poseidon2 hashes, while signatures remain classically verifiable by anyone against published node keys. We consider stating this boundary precisely to be part of the product.

Beyond decisions.
The full agent accountability stack.

Modern AI systems are not single-shot decision engines — they are autonomous agents that reason across multiple steps, call external tools, maintain state across interactions, and delegate to other agents. Rubric attests the entire stack, not just the final output.

Tool Call Attestation

Every external tool invocation — web search, code execution, API calls, database queries — is attested with hashed input/output pairs, duration, success status, and chain linkage to the prior attestation. This is where AI risk actually lives: not the reasoning, the action.

Agent State Hashing

At every decision point, Rubric captures a SHA3-256 hash of the agent's current memory, context window, and system prompt — capped at 4KB to bound storage impact. Two identical decisions made from different mental states produce different state hashes, making behavioral drift cryptographically detectable. State hash type is recorded as full or truncated for audit transparency.

Chain Linking

Every attestation carries a priorAttestationId reference, forming a tamper-evident chain across the agent's full session. Deleting or reordering any attestation breaks the chain. The complete reasoning sequence — from initial input through every tool call and state transition to final decision — is cryptographically sealed and independently verifiable.

Agent Identity Registry

Each agent is registered with a stable W3C DID anchored to Hedera mainnet and a dedicated ML-DSA-65 (NIST FIPS 204) keypair. Agents are first-class cryptographic principals — auditors can verify which agent acted, not just that an agent acted. Registration records include jurisdiction, permitted uses, model hash, and version.

Global regulatory compliance.
Computed automatically from intent.

Declare your AI system's intended use and jurisdiction. Rubric computes your complete regulatory obligation set automatically — mandatory fields, applicable articles, safe harbor conditions, and a ready-to-paste SDK configuration. No compliance expertise required.

Every assertion in every compliance report is backed by a cryptographic attestation anchored to Hedera HCS Topic 0.0.10416909 — independently verifiable without Rubric's involvement. Regulators, notified bodies, and auditors can verify any evidence entry directly on-chain.

17 frameworks.
Four packages. Zero configuration.

Rubric ships as four distribution packages across Python, TypeScript/Node, and MCP — covering 17 live framework integrations. Installation takes under 60 seconds. Every package includes the full compliance field set and auto-populates timing, latency, and event classification.

Tested at scale.
Validated numbers.

Your existing tools.
No custom integration.

Every attestation is exportable on demand, in the format your stack already speaks. Pipe directly into your existing GRC or SIEM platform — no custom integration, no professional services engagement.

For Splunk environments, the Rubric Protocol Add-on for Splunk is published on Splunkbase (app 9176, approved July 2026 following Splunk AppInspect review) — field extractions for attestation events delivered as JSON or CEF via HTTP Event Collector, compatible with Splunk Enterprise and Splunk Cloud, platform versions 9.0–10.5. Each parsed event carries a verify_url to Rubric’s public verifier for independent confirmation.

Each export record includes a verify_url linking to the public audit portal — regulators can independently verify any attestation without access to your internal systems or Rubric infrastructure.

Real-time event push.
Every decision, instantly.

Every attestation can be pushed to your SIEM, GRC platform, or custom endpoint in real time via HMAC-SHA256 signed webhooks. Non-blocking — entirely outside the attestation pipeline's request flow.

Webhooks auto-disable after 10 consecutive delivery failures — preventing noise while surfacing endpoint issues. Delivery status and failure counts are tracked per endpoint and visible via GET /v1/webhooks.

Start free.
Scale when you do.

Design partner pricing available: 60-day free pilot followed by $2,500/month. Contact Scott@Rubric-Protocol.com to discuss requirements.