The law
Europe’s AI law, in plain words.
What the law asks, and the three duties Atrytone is built around.
Europe's law on artificial intelligence, adopted in 2024, its duties applying in stages from 2025 to 2028 (the European Commission's timetable, read October 2026). It requires human oversight and record-keeping of high-risk systems, and transparency of systems that write content or talk to people. Atrytone is built around all three.
HUMAN OVERSIGHT · ART. 14a person signs off; nothing irreversible without askingRECORD-KEEPING · ART. 12every act, decision and cost on a ledger an auditor can readTRANSPARENCY · ART. 50every commit and pull request that carries its work says AI models made itOur provisional reading treats Atrytone as a limited/minimal-risk AI system. It was taken on 1 September 2026 and confirmed by the operator on 4 October 2026. No lawyer has reviewed it yet, so we do not claim it.
What we do not claim
Atrytone carries no CE mark.
What is still open
- Whether the duties change with what you build. If the software you make with Atrytone is itself closely regulated, such as a tool for hiring, credit or essential public services, the duties on you and on us may be different. Ask us before you start such work.
- Whether our reading holds. No lawyer has yet assessed how the law applies to Atrytone, or to the way you use it. Until one has, nothing on this site says a legal duty is met, only what the product does.
Technical details
Who is who
The companies that make the AI models, such as Anthropic, OpenAI and Google, are the providers of those models, and the duties the law sets for general-purpose models are theirs. Atrytone is offered under its own name and puts those models to work on your code: we are the provider of Atrytone, the system, not of the models inside it. You, using Atrytone in your work, are its deployer.