Grades

A grade names one degree of AI involvement and carries the facts the rest of the module reads: a severity for ordering, the EU icon variant to show, the disclosure sentence the reader sees, and whether AI generated or manipulated the published content at all. Grades are config entities. Manage them at /admin/config/content/ai-disclosure/grades.

A grade says how AI was involved. It does not say whether Article 50(4) applies: that is the scope assessment, answered on the content and on the profile.

The nine shipped grades

Machine name	Label	Pick it when	Severity	EU icon	Label required
human_only	Human only	No AI was involved	0	none	No
ai_metadata	Metadata picked by AI	AI only chose tags, categories or similar	10	none	No
ai_rated	Rated / scored by AI	AI only scored or ranked something	20	ai	No
ai_translated	AI translated	AI translated the text	30	ai_modified	Yes
ai_summarized	AI summarized	AI produced a summary	40	ai_generated	Yes
ai_partly_assisted	Partly AI assisted	AI wrote or changed part of it	50	ai_modified	Yes
ai_assisted_hitl	AI assisted (human led)	A person led, AI helped throughout	60	ai_modified	Yes
ai_generated_autonomous	Autonomously AI generated	AI produced it, no person writing	70	ai_generated	Yes
ai_deepfake	AI deep fake (media)	AI made image, audio or video depicting real people or events	80	ai_generated	Yes

Label required is the factual premise both paragraphs of Article 50(4) share: with this grade, AI generated or manipulated the content that was published. The first three do not meet it and fall outside both paragraphs, so they never need an assessment.

Grades used to carry a second flag, review_exemptible, and ai_deepfake was the only one set to No. It was standing in for something a grade cannot know: whether the content is a deep fake or text. That is now an explicit answer on the content, so the flag is gone.

Severity orders the grades and drives the recorder's never-downgrade rule: an automated caller can only raise the effective grade, never lower it.

Each grade also stores the IPTC digital source type URI that feeds the machine-readable marking. The field is optional, and what it accepts is either nothing or an absolute URI such as http://cv.iptc.org/newscodes/digitalsourcetype/trainedAlgorithmicMedia. The module refuses anything else wherever it comes from, the form, a save from code, or a config import, because the value is published as it stands.

The machine names are a stable API

Other modules reference grades by ID, so the shipped machine names are not renamed within a major version. Everything else on a grade is editable by the site: the label, the sentence, the severity, the icon, and the legal flag.

Before editing Label required, read Article 50. The shipped configuration states whether AI generated or manipulated the published content. Editing it changes that statement, not only its wording.

Visible parts

Each grade also carries six checkboxes (icon, sentence, legal line, description, editorial statement link, editorial responsibility) that decide which parts of the rendered disclosure show for content at that grade. They only switch a part on or off. They never move where its data comes from: the legal line and the statement link still come from the site's Policy tab, the description still comes from the profile or the item, and editorial responsibility still comes from whoever set it on the profile or the item.

Grade	Icon	Sentence	Legal	Description	Statement	Editorial responsibility
human_only	No	No	No	No	No	No
ai_metadata	No	Yes	No	No	No	No
ai_rated	Yes	Yes	No	No	No	No
ai_translated	Yes	Yes	Yes	No	Yes	No
ai_summarized	Yes	Yes	Yes	Yes	Yes	No
ai_partly_assisted	Yes	Yes	Yes	Yes	Yes	No
ai_assisted_hitl	Yes	Yes	Yes	Yes	Yes	No
ai_generated_autonomous	Yes	Yes	Yes	Yes	Yes	No
ai_deepfake	Yes	Yes	Yes	Yes	Yes	No

When Label required resolves to Required, the icon and the sentence render whatever the checkboxes say: Article 50(4) does not bend to a display preference. Nothing else is exempted: unchecking the legal line, the description, the statement link or editorial responsibility still hides them, even on a required label.

Editorial responsibility ships off on every grade, unlike the other five: the value it shows can be a natural person's name, and naming one is the site's own choice to publish, not something switched on for it. Turning it on for a grade shows the name or team a site has recorded, on the profile or the item, as responsible for that content. Leaving it off never hides the rest of the disclosure.

An already-installed site keeps every one of the other five parts on until it runs the update that sets this table on the nine shipped grades: those properties' own default is "everything on", so nothing changes for a site that never runs it, and it never touches a grade the site created itself. After the update, a site that relied on ai_translated always showing its description, for example, stops seeing it, and gets it back by checking that grade's "Description" box again. Editorial responsibility carries no such transition: its own default is off everywhere, upgrading site and new install alike.