EU AI ACT · ARTICLE 13 · PATENT PENDING

Evidence that can support an Article 13 review.

Hive can create signed records for selected AI actions. The exact fields, algorithm, verifier, and retention period depend on the selected endpoint and receipt profile. This supports Article 13 evidence requirements, subject to system classification, deployment controls, and counsel review.

Article 13 · Annex IV

Selected evidence needs mapped to available features.

This capability map can support an Article 13 evidence design. It is not a legal conclusion or a claim that any deployment is compliant.

REQ-01

Log what model ran

When the selected integration and receipt profile capture these fields, the record can bind the model identifier, provider, and version to the action.

model_id · provider · version
REQ-02

Log inputs and outputs

When the selected integration and receipt profile capture them, token counts and prompt and completion hashes can support review without storing raw content.

token_counts · prompt_hash · completion_hash
REQ-03

Tamper-evident records

A supported receipt profile can bind selected fields with the algorithm named by that endpoint. The relying party must use the matching verifier and acceptance policy.

profile-specific schema · named algorithm
REQ-04

Portable audit trail

Supported receipt profiles use portable JSON. Offline verification is available only where the selected profile publishes a compatible verifier and trust key.

portable JSON · profile-specific verifier
REQ-05

Long-horizon retention

ML-DSA-65 is a NIST-standardized post-quantum signature option. Long-term verification also depends on retained artifacts, keys, formats, validation software, and migration policy.

NIST FIPS 204 · quantum-safe
REQ-06

Third-party verifiability

The public /verify page currently checks the supported Ed25519 mesh receipt and countersignature forms. Typed, ML-DSA, and hybrid envelopes require their named compatible verifier.

scope stated beside each verifier
Competitive analysis

How Hive compares

Side-by-side with KITE AI and conventional logging approaches.

Feature Hive KITE AI Status quo (logs)
Signing scheme ML-DSA-65 (NIST PQ) ECDSA only None
Receipt portability Self-contained, offline Chain lookup required N/A
Inference provenance Model + provider + latency signed None None
ZK selective disclosure Aurora Lattice None None
Hardware root of trust HiveSeal (Patent Pending) None None
EU AI Act posture Supports evidence design Not addressed Not addressed
Chain dependency None, Base optional Kite L1 required None
Open verifier Yes, for named supported forms No No
Post-quantum signature option Profile-dependent No No
How it works

Three steps from inference to verified audit trail

No changes to your agent logic. No new infrastructure to operate.

1

Agent makes inference call

Hive catches the call and records model identity, provider, token counts, and latency, all before the response reaches your agent.

model_id · provider · tokens · latency_ms
2

Apply the selected receipt-profile signature

The selected receipt profile binds its declared fields with its named algorithm. Production acceptance depends on the published key, compatible verifier, and agreed profile.

declared schema · named algorithm · trust key
3

Auditor verifies instantly

An auditor uses the verifier that matches the selected receipt profile. The public /verify page states which Ed25519 forms it accepts and which forms it does not.

matching verifier · stated acceptance limits
Compliance coverage

Frameworks covered

Hive PQ Receipts and Inference Provenance address obligations across multiple regulatory and standards frameworks.

EU AI Act Art. 13 GDPR Art. 5(1)(f) DORA Art. 17 NIS2 SOC 2 Type II ISO 27001 HIPAA FinCEN
Enterprise · Article 13

Design the evidence before deployment.

Hive supports Article 13 evidence requirements, subject to system classification, deployment controls, and counsel review. The order form must name the receipt profile, fields, verifier, retention period, custody model, and any optional chain anchor.

Patent Pending · Evidence support only. Classification, controls, and legal conclusions remain deployment-specific.

Data handling. Client-side hashes and server-side text processing have different boundaries. Logging, storage and retention are endpoint-specific. See Privacy.