EN 18286:2026: the *logging standard* high-risk AI systems will need

What the new standard says

EN 18286:2026 is the European standard on automatic logging for high-risk AI systems ^[1]. It tells you what events to log, how often and for how long. The standard applies to systems that fall under Annex III of the AI Act, which covers areas like biometric identification, critical infrastructure, employment, credit scoring and law enforcement ^[2].

Think of it like a car's service book. The standard says which checks to record and when, but it doesn't say how you prove nobody tore out a page.

Article 12 of the AI Act requires high-risk systems to technically allow for the automatic recording of events over their lifetime ^[3]. EN 18286 fills in the detail. It covers input data references, model versions, decisions, timestamps and who reviewed what. It doesn't, however, say how to protect the logs once they exist.

The five trustworthiness standards

The logging standard is one of five. The original framework, prEN 18229, has been split into separate parts ^[1]:

• EN 18286 on logging,
• standards on transparency, human oversight, accuracy and robustness.

Each one addresses a technical requirement from the AI Act. Together, they form the practical rulebook for compliance. The standards are expected to be adopted in December 2026, roughly a year before the Annex III high-risk requirements apply on 2 December 2027 ^[2].

What it means for organisations running high-risk AI

If your AI system falls under Annex III, you need automatic logging from December 2027. EN 18286 tells you what to log. What it leaves open is how you show an auditor, a supervisor or a court that a log hasn't been quietly edited after the fact.

Logs are like receipts. Nobody asks to see them until there is a dispute. When that happens, a log kept by the same organisation that benefits from what it says is easier to challenge than one that can be independently verified.

Access controls and monitoring are the foundation, but they don't solve the last-mile problem: the people who keep the logs could also change them. An independent fingerprint per log entry goes one step further, because a change shows up even when it is made by someone with the right access.

How a digital witness fits

Here, the thing to seal is each log entry as it is written: the input reference, the model version, the output, the timestamp and the reviewer. Sealing takes a fingerprint of the entry, not the entry itself, so the log never leaves your systems.

The fingerprint is sealed on WitniumChain, where any later change would show, even one made by us. To check, anyone you share the log with calculates the fingerprint again and compares it with the sealed one. If it matches, the entry is exactly what was written at that time. Change a single character and it won't match.

Auditors, supervisors and, if it comes to it, a court can check the log themselves, without having to trust whoever kept it.

What it doesn't do: it doesn't show that the decision was fair, lawful or correct, or that the log is complete. It shows that the entries you sealed haven't changed.

How it would work

1. Your AI system writes a log entry and sends a fingerprint of it to WitniumChain through the REST API, the SDK or the MCP server.
2. WitniumChain seals the fingerprint and returns a witness ID.
3. When the log is questioned, anyone you share it with can check that the entries match the sealed fingerprints.

The content stays with you. The first 2,000 witnesses a month are free. Read the docs or get access now.

What this means for you

If you are building or running a high-risk AI system:

• Check whether your system falls under Annex III. If it does, you have until 2 December 2027 to meet the logging requirement.
• Read EN 18286 when it is published in December 2026 and map it to your system: which events you log, in what format and for how long.
• Decide how you will show that your logs haven't been tampered with. Access controls and monitoring are necessary, but they don't stop someone with the right access from editing a log.
• Ask your compliance team whether an independent fingerprint per entry makes sense for your system.

What to watch

The five trustworthiness standards are expected to be adopted in December 2026 ^[1]. Once they are published, organisations have roughly a year to implement them before the Annex III high-risk requirements apply on 2 December 2027 ^[2].

For systems under Annex I, the requirements apply from 2 August 2028 ^[2].