EN 18286 is published: the first piece of the AI Act standards puzzle

The first standard is here

EN 18286 arrived in October 2024, seven months before the main AI Act takes effect ^[1]. It's the first harmonised standard under the regulation, and it tackles Article 17: quality management systems for providers of high-risk AI.

Think of it like getting the assembly instructions before the furniture arrives. For anyone building or selling high-risk AI, this standard gives you the checklist the regulation has been promising. Comply with it and you get a presumption of conformity with Article 17, meaning authorities assume you've met that part of the law unless proven otherwise ^[3].

What the standard covers

Article 17 requires providers of high-risk AI systems to put in place a quality management system ^[4]. EN 18286 spells out what that means in practice: policies, procedures and documentation covering the entire lifecycle of the system ^[1].

The standard addresses data governance, record keeping, resource management, design and development controls, post-market monitoring and incident reporting ^[1]. It is written to work alongside existing quality frameworks, so if you already have ISO 9001 or similar in place, EN 18286 builds on that rather than replacing it ^[1].

One detail worth noting: the standard requires automatic logging of events, which ties directly to Article 12 of the AI Act ^[5]. Logs are what tell you what your AI did on a given day. The standard tells you to keep them. What it doesn't say is how you prove nobody has edited them afterwards.

Who it applies to, and when

EN 18286 applies to providers of high-risk AI systems listed in Annex III of the AI Act ^[2]. That includes AI used for biometric identification, management of critical infrastructure, recruitment, credit scoring for individuals, pricing of life and health insurance, emergency triage and law enforcement.

The deadline is 2 December 2027 ^[2]. After the Digital Omnibus amendments, that is when Annex III high-risk systems must comply with the full set of requirements, including Article 17.

If you provide one of those systems, you have just over a year to get your quality management sorted. The standard is voluntary, but it offers the clearest path to showing compliance.

What this means for you

If you're building or selling high-risk AI, here are three questions to ask now:

• Do you have a quality management system that covers the AI lifecycle, from design to post-market monitoring?
• Are you logging the events Article 12 requires, and can you show those logs haven't been tampered with?
• If you already have ISO 9001 or a similar framework, where are the gaps EN 18286 would fill?

For those waiting to see what compliance looks like in practice, this standard is the first concrete answer. Around 40 more are expected, covering areas like data quality, transparency and human oversight ^[1]. EN 18286 is the starting block.

Logs you can check

Here, the thing to seal is each log record as it is written: the event type, the timestamp, the system state and any relevant output. Sealing takes a fingerprint of the record, not the record itself, so the data never leaves your systems.

If an auditor or a supervisor asks to see your logs six months later, anyone you share them with can check that each record is exactly what was written at the time. The check works by calculating the fingerprint again and comparing it with the sealed one. If they match, nothing has changed.

What it doesn't do is prove the logs are complete, or that the decisions they record were correct. It shows one thing: that the records you're presenting now are the ones written then, unchanged.

You can seal log records as they are written, or in batches every few minutes, through the REST API, the SDK or the MCP server at WitniumChain. The content stays with you. Auditors, supervisors and, if it comes to it, a court can check the fingerprints. Read the docs for the details. The first 2,000 witnesses a month are free.

What to watch

The European Commission's standardisation request to CEN and CENELEC, issued in February 2023, lists the remaining areas for which standards are being developed ^[6]. Expect standards on data quality, transparency, accuracy, robustness and human oversight over the coming months.

For Annex III systems, the compliance date is 2 December 2027 ^[2]. For general-purpose AI models with systemic risk, certain obligations applied from 2 August 2025 ^[2]. Mark the dates, read the standards as they arrive, and decide early which parts of your quality management need work.