In August we wrote about four AI labels: AI generated, AI modified, AI assisted, human authored. That post is still accurate for what it covered. It is no longer the full picture. Content Label Terms v1.2, published 21 September 2026 by Philip Kübler, CEO and attorney at law at ProLitteris, sets out six labels with binding definitions, not four with informal ones. The two additions, HUMAN AUTHORED and HUMAN APPROVED, are not a rename of what the earlier post called "human authored". They are a different kind of label, and the difference is the reason this post exists.
Four labels describe the content. Two describe the person behind it
The four AI Labels, AI FREE, AI ASSISTED, AI MODIFIED and AI GENERATED, describe how a piece of content was produced. They answer "what happened here." The two Human Labels, HUMAN AUTHORED and HUMAN APPROVED, describe human creativity and human review, respectively. They answer a different question: "who stands behind this, and can that be verified." That second question is why the two Human Labels are identity-bound in a way the four AI Labels are not: using them means the natural person making the declaration has gone through a qualified identity check, a passport-based process provided by Swisscom Trust Services.
The four AI labels, one rule each
Exactly one of these applies to a given piece of content and version. Never zero, never two.
AI FREE. Content without a defining AI contribution: generated entirely by humans or machines, without AI, or with only minimal technical, preparatory or checking use of it. A text with or without machine spell-checking. A photograph with or without simple lighting correction. An unmodified smartphone video.
AI ASSISTED. Hybrid content with a defining AI contribution alongside defining human or machine contributions, generated predominantly by humans, with at least one defining element generated or modified by AI. A human-written text in which AI drafted a defining passage or a heading. An article where a person decided the selection and arrangement and AI contributed defining text.
AI MODIFIED. Content with a defining AI contribution built on pre-existing content that stays recognisable, with at least one defining element modified by AI. An existing photograph into which objects or people have been inserted or removed with AI. A speech recording whose spoken words have been changed with AI.
AI GENERATED. Content created predominantly by AI, entirely or in its defining elements. Output from a prompt or a chat: text, images, music, voices or video. A draft produced by AI and used without further defining human editing.
Two rules for the doubtful cases
Two rules exist for exactly this reason: teams reach for the softer label when the honest one is uncomfortable. Where it's unclear which of the four applies, the stronger disclosure wins, AI GENERATED over AI ASSISTED, AI ASSISTED over AI FREE. And where content is built on something pre-existing and still recognisable after an AI change to a defining element, AI MODIFIED is the label, not AI ASSISTED, even if the change feels minor. The label also travels with the content. It doesn't reset depending on where or how the piece is published.
HUMAN AUTHORED: the identity-bound label for AI FREE content
HUMAN AUTHORED is not a fifth AI label. It's an additional, identity-bound designation that can only sit on top of AI FREE. It cannot combine with AI ASSISTED, AI MODIFIED or AI GENERATED, by definition, not by convention. Declaring it means an identified person is attaching their name to "I made this, without a defining AI contribution," in a way that can be checked back against a verified identity later.
HUMAN APPROVED: review and responsibility, by a different person
HUMAN APPROVED works differently and is more flexible. It can combine with any of the four AI Labels, including AI GENERATED. It says an identified person reviewed and finalised the complete content and takes responsibility for it, regardless of how much AI was involved in producing it. The one hard rule: where a content already carries HUMAN AUTHORED, adding HUMAN APPROVED requires a review and approval by a different natural person, not the same one wearing two hats. For the same content and version, one person may hold only one of the two Human Labels, not both.
Which one does a newsroom, agency or studio actually need
Most working content isn't AI FREE from an individual working alone, and most of it isn't AI GENERATED with no human editing either. A publisher running an AI-assisted draft through an editor's sign-off before it goes out is a HUMAN APPROVED case on top of AI ASSISTED or AI GENERATED, not a HUMAN AUTHORED one. A columnist writing entirely without AI, willing to put their verified name behind that specific claim, is the HUMAN AUTHORED case. The two labels aren't interchangeable ways of saying "a person was involved." They answer different questions, for different situations, and picking the wrong one is a less defensible position than picking the accurate one.
Where this sits next to the AI Act, not because of it
Since 2 August 2026, Article 50 of the EU AI Act has required disclosure that certain AI-generated or AI-manipulated content is artificial. That's a transparency obligation on providers and deployers under EU law. We cover what Article 50 actually asks for on our Article 50 page. The six Content Labels are a separate thing: ProLitteris and Swiss Trust Layer's own trust-infrastructure system for declaring how content was made and who stands behind it. The Labels sit next to the Article 50 obligation as one way a team might satisfy part of it in practice; they aren't the legal requirement itself, and using them isn't a claim of regulatory certification.
A label is a statement. A sealed one is a statement you can still prove
Every Content Label is a declaration a User makes to the best of their knowledge, on a sufficient factual basis, for content they're authorised to declare for. That's true whether it's written on a page or attached inside a trust infrastructure. The difference is what happens when someone asks you to support it six months later. A label typed onto a webpage can be edited afterwards, including by you. A declaration tied to the exact file, fixed to a specific version, and carried alongside a qualified timestamp, can't quietly change after the fact. We cover how a sealed record ties a declaration to a specific file and a verified identity in the MyCopyright register overview.
Pick the label that's accurate. Then make it one you can still stand behind later.
