For organizations deploying high-risk AI systems in the EU market. A technical and procedural mapping of how AXIS delegation chains satisfy the record-keeping, traceability, and accountability requirements that enter full application on 2 August 2026.
The EU AI Act enters full application on 2 August 2026. For any high-risk AI system placed on the EU market, Article 12 is the provision that will hit operations hardest.
Article 12 requires high-risk AI systems to "technically allow for the automatic recording of events (logs) over the lifetime of the system." The regulatory text is short. The implementation is not. Logs must establish traceability for three distinct purposes: risk identification under Article 79(1), post-market monitoring under Article 72, and operational oversight by deployers under Article 26(5). Logs must be retained for at least six months (Article 26(6)) and made available to market surveillance authorities on request.
The standard most compliance teams are building toward goes further than the regulation's bare text. Regulators and litigators increasingly expect logs that can reconstruct the who, what, when, and why for every decision and exception and that trace every action to a named, accountable human. A server log timestamp is no longer sufficient.
This is where most current AI deployments fail. Not because they don't log, but because their logs can't reconstruct a chain of authority. An AI agent's log entry says "action taken at timestamp X, using model Y." It does not say "under delegation from employee Z, authorized within scope S, verifiable by signature V." When a regulator asks "under whose authority did this agent act," the answer is usually a Slack message, a prompt template, or a vendor contract. None of those are tamper-evident. None of those survive a hostile audit.
Five elements are consistently missing from current AI system logs:
1. Attribution to a living person. AI agent actions are typically attributed to a service account or a model name. Article 12 compliance, as interpreted by emerging standards like EN 18229-1 and ISO/IEC 24970, increasingly expects a direct trace to a human principal.
2. Authority scope at time of action. Logs record what the agent did. They rarely record what the agent was authorized to do at the moment it acted. The difference matters when an action falls outside scope.
3. Delegation chain integrity. When AI work crosses operator boundaries (a vendor's AI acting on a customer's platform, a multi-agent workflow spanning services), the chain of who authorized what is usually implicit or contractual. It is not cryptographic.
4. Revocation state at time of action. When a delegation is revoked, current systems rely on eventual consistency. An agent can act on a stale authorization for minutes or hours.
5. Cross-system audit reconstruction. Agent actions routinely span multiple systems: the AI tool, the orchestration layer, downstream SaaS, the model provider. Logs live in five places in five formats.
Organizations deploying high-risk AI systems need to close all five gaps by 2 August 2026. The penalty for failing Article 12 is up to €15 million or 3% of global annual turnover, whichever is higher.
AXIS is an open protocol for agent identity, delegation, and authorization across operator boundaries. Its core artifact, a signed delegation credential carrying a cryptographic chain from human principal to acting agent, maps directly to the five gaps above.
Every AXIS-signed agent action produces an artifact with five properties Article 12 requires:
| Article 12 requirement | What AXIS provides |
|---|---|
| Automatic recording of events over system lifetime (Art. 12(1)) | Every delegated action produces a signed delegation credential and an action record anchored to it. Credential signing is automatic as part of the call pattern. |
| Traceability for risk identification under Art. 79(1) (Art. 12(2)(a)) | Delegation chain shows the full authority path. Actions outside scope produce verification failures, which are themselves logged. |
| Post-market monitoring (Art. 12(2)(b), Art. 72) | Audit records anchored to AXIS-signed credentials can be exported to monitoring pipelines, and the credential signatures stay independently verifiable. |
| Deployer operational oversight (Art. 12(2)(c), Art. 26(5)) | Deployers receive the same cryptographically signed credentials the provider produces. No trust gap between operators. |
| Minimum 6-month log retention (Art. 26(6)) | Records are plain JSON artifacts anchored to signed credentials. Standard log retention systems support them without modification. |
| Tamper-evident records | Ed25519 signatures on the delegation chain detect any modification of the credential. Verification is computationally trivial. |
| Traceable to a living person | The root of every delegation chain is a human principal's signing key, bound to an operator identity record. |
Article 12 is not the only relevant provision. Several other Articles touch on capabilities AXIS directly supports:
Being clear about scope matters. AXIS is accountability and identity infrastructure. It does not address several distinct obligations under the Act.
"AXIS supports Article 12 compliance" is an accurate claim. "AXIS makes you EU AI Act compliant" is not. Compliance requires an integrated program with documentation, controls, and assessments across many Articles. AXIS is one component of that program, serving the record-keeping, traceability, and delegation layer specifically.
For organizations evaluating AXIS as part of EU AI Act readiness:
For organizations building high-risk AI systems into the EU market:
If the answer to any of these is uncertain, the gap is material before 2 August 2026.
This document describes how the AXIS protocol supports Article 12 compliance. It is not legal advice. Compliance with the EU AI Act requires a comprehensive program assessed against all applicable Articles. Consult qualified counsel for a compliance determination specific to your systems and jurisdiction.