Which tools are AI systems under the EU AI Act.
Cernyx documents, for every tool used across the entertainment industry, whether it contains an AI system under the EU AI Act, at which risk tier it sits, and how it performs against the seven requirements for trustworthy AI. Structured, sourced, and recorded per component, not per product name.
Tool vendors
Components
Are AI systems
Sectors in scope
What Cernyx answers
Three questions, and the reasoning behind each answer.
Art. 3(1)
Is it an AI system?
A tool counts as an AI system only when a learned, inferring model produces its outputs. Deterministic, rule-based logic authored by people does not, however it is branded.
The AI Act ladder
At which risk tier?
Each AI system is placed on the Act’s ladder: minimal, limited with Art. 50 duties, Art. 50 plus GPAI, high-risk under Annex III or, rarely, prohibited under Art. 5. Where the tier depends on an open point, such as configuration or deployer use, the most severe plausible tier is recorded until research settles it. The route and its legal basis are recorded, not just the label.
7 HLEG requirements
How does it perform on trustworthy-AI?
Each tool is scored against the seven HLEG requirements. Where a requirement presupposes an AI system the tool has none, it is marked not applicable rather than scored zero, so an inapplicable duty is never misread as a failure.
Sectors in scope
The entertainment industry
The research began with games, the most tool-dense pipeline, and is extending across music, film, and live entertainment.
Games
The sharpest test of the legal definition. 562 classifications to date.
Music
Generative audio, mastering, and rights tooling under active classification.
Classification in progress
Film & video
Editing, VFX, and post-production pipelines mapped against the Act.
Classification in progress
Dance & live
Staging, projection, and live-performance systems, scoped for the next phase.
Scoping
The gap Cernyx fills
Why this is difficult to establish from existing sources.
The EU AI Act applies across every industry that ships software. Studios, publishers and platforms now have to know which tools in their pipeline carry obligations, and which do not. Three things make that difficult.
01
Marketing obscures the legal reality
“AI” is a selling word. A tool with “.ai” in its name may be a deterministic lookup service; a plain-sounding library may embed a learned model. Neither the marketing nor a quick search reliably tells a studio whether a tool is an AI system under the Act.
02
The obligation sits per component
The Act classifies systems and techniques, not product names. “Unity uses AI” or “Photon is just netcode” tells a studio nothing actionable. The answer sits one level down, at the component: one specific version of one specific tool, assessed on its own.
03
Existing sources aren't independent
Vendor self-declarations are interested parties. AI news is not legally structured. Legal commentary is expensive and case-by-case. What has been missing is a neutral, sourced, industry-wide register.
The games industry research
Where obligations sit, and where they do not.
Snapshot v55 · 562 classifications
Roughly half of the tools studios reach for are not AI systems at all, a defensible exemption on a compliance filing. The high-risk surface is small and specific, which is exactly what lets enforcement focus where it belongs. Three of the six high-risk classifications, and the single prohibited one, describe an industry practice rather than a specific product. The prohibited classification covers biometric categorisation for personalisation under Art. 5(1)(g), recorded pending research into which attributes are inferred.
Independent by design
A classification that does not depend on trusting any one party.
Cernyx is a product of AI Distribution B.V., which builds and maintains the register. It is presented as an independent platform, structurally distinct from AI Distribution, for one reason: the judgement must stand on the evidence and the law, not on the brand behind it.
AI Distribution is never the approving party. Each classification can carry two independent signatures.
The independent reviewer
A Certified AI Compliance Officer, a Certified AI Business Leader, or an external AI-specialised lawyer, always independent of the vendor, certifies the judgement.
The vendor's accountable owner
The vendor’s own accountable owner certifies that the facts on which the classification rests are true. AI Distribution builds and runs the register. It is not the scorekeeper.
The register is public and sourced.
A per-component classification for the tools that are exempt, and the Article 50 duties for the ones that are not.