Which AI Act risk class is your system in?
| Rules | AI Act as amended in 2026 |
|---|---|
| Questions | Up to 8 |
| You get | Risk class, controls, dates |
| Rules as of | 12 November 2026 |
The check
Answer the questions, then read your result
Dates
What applies when
Prohibited practices apply
General-purpose model duties apply
Transparency duties apply to new systems
Tell people they are talking to AI; label deepfakes; mark generated content.
Marking deadline for systems already on the market
Also new bans on nudifier apps and generated abuse material.
High-risk duties for Annex III systems
Hiring, credit, insurance pricing, education, essential services, biometrics.
High-risk duties for AI in regulated products
Medical devices, machinery, vehicles and other Annex I products.
Sources
Rules as of 12 November 2026
- Digital Omnibus on AI: amended dates (Cooley)
- AI Act amendments adopted under the Digital Omnibus (Loyens & Loeff)
- Regulation (EU) 2024/1689, the AI Act (EUR-Lex)
This is an engineering view of public rules, not legal advice. Your counsel and your regulator decide what applies to you.
We update the tool when the rules or prices change and show the date above.
Go deeper
Related pages and tools
- AI harness
Agents that run inside a controlled harness, with a named person signing every release.
Capability - AI governance
Logging, review and evaluation built into delivery.
Capability - Evidence workload check
What gathering evidence by hand costs your team.
Tool - Clinical Research Harness
An audited agent harness for clinical programming.
Capability
Next step
Build the controls once, not per audit
A 30 minute call with Oleg. We look at your system and agree which controls belong in the first release.
Oleg Pun
CEO, Vareger
