Framework / EU AI Act

High-risk AI must keep logs.
Article 12, for your agents.

For high-risk systems, the EU AI Act requires automatic record-keeping (Article 12) and incident reporting (Article 73). AIR produces the Article 12 log for the agent layer — signed, attributable — and is ready when Article 73's clock starts.

Article 12Article 73high-riskAnnex III

Book an agent audit See the evidence ↓

Automatic logging · incident-ready · tamper-evident

Article 12 log · live

Loggedagent action · automatic

Attributedagent + authorizing human

Art. 73 readyreconstruct in minutes

Out-of-scopeblocked · SV-SCOPE

signed in-process · anchored Rekor

✓ lifetime log · tamper-evident

01The stakes

Annex III high-risk · agent system

Your high-risk system runs on agents. Article 12 wants the logs; Article 73 wants the incident report.

Article 12 requires automatic logging over the system's lifetime, and Article 73 requires reporting serious incidents on a deadline. AIR keeps the log continuously, and reconstructs the incident in minutes when the clock starts.

02The articles

Two obligations, and an honest date.

Article 12

record-keeping

High-risk systems must automatically record events over their lifetime. AIR signs each agent action in-process — the Article 12 log, for the agent layer.

Automatic, not reconstructed after the fact.

Article 73

incident reporting

Serious incidents must be reported on a deadline. The causal graph and replay reconstruct what happened, fast, with a signed record.

Minutes, against days of cross-system log-spelunking.

The timeline

Digital Omnibus

The Digital Omnibus (provisional agreement May 7, 2026) defers high-risk Annex III obligations from Aug 2, 2026 to Dec 2, 2027.

Provisional until adopted and published in the Official Journal; until then Aug 2, 2026 stays active. Article 5 prohibitions apply since Feb 2025.

03How AIR answers

What the high-risk obligations ask of the agent layer.

Automatic logging over the system's lifetime? (Art. 12)

ProveSigned in-process, every action, retained and re-anchored.

Who and what acted?

AccountEach action bound to a named human or service.

Report a serious incident on deadline? (Art. 73)

InvestigateCausal graph and replay reconstruct it in minutes.

Tamper-evident for a regulator?

ProveAnchored to a public transparency log.

Across cross-agent handoffs?

HandoffA chain of custody survives delegation.

04The evidence

The Article 12 log, and the Article 73 report under it.

An automatic, signed, attributable record of every agent action over the system's lifetime — and, when an incident hits, the reconstruction that backs the Article 73 report, admissible under FRE 902(13)–(14).

See all framework mappings →

Article 12 log · agdr/v2

Logged · agent action — automatic, time-stamped

Attributed · agent + authorizing human

Art. 73 · reconstructable in minutes

Halt · out-of-scope call blocked — SV-SCOPE

signed in-process · blake3 · ed25519 · anchored Rekor

✓ lifetime log, tamper-evident · search.sigstore.dev

05What you get

The tiers high-risk providers choose.

Enterprise

most teams here

Article 12 logging for the agent layer
Article 73 incident reconstruction
Causal graph, query & replay
SIEM: Splunk · Datadog · Sentinel · Sumo
SSO / OIDC, SLA

Book an agent audit

Air-gapped

regulated · sovereign

Everything in Enterprise, plus

On-prem / offline anchoring
No phone-home — records never leave
Extended retention + PQ re-anchoring
Admissibility Pack — FRE 902 + expert support

Talk to us

Be ready for Article 12, before the date moves again.

A free agent audit shows the lifetime log you'd already have for your high-risk agents — and how fast you could file under Article 73.

Book an agent audit →