Defining publication conditions for AI-generated profile images

publication and legal assessment classified confirmed facts, unconfirmed facts, and prohibited inferences across the intended Japanese and English copy, AI personas, profile images, claims, calls to action, and locale display. It explicitly did not provide final legal advice, jurisdiction-specific applicability, final legal approval, or publication approval.

The CEO confirmed that the intended profile images were AI-generated and had no objection to the direction of the existing Japanese and English disclosures. No copy change was requested. The retained disclosure direction contained three elements: AI Talent represents role-bearing AI personas; they are not humans, real people, or human employees; and the portraits are AI-generated representations of those personas.

• Place the full locale-matched disclosure in ordinary reading order before the first image in the same profile collection.
• Do not substitute hover, collapsed content, a tooltip, image alternative text, or a Legal page alone for the notice.
• When an image is reused away from the collection, such as in a Hero, card, modal, search image, or social image, provide an equivalent nearby disclosure in that context.
• Keep the notice readable at responsive widths and ordinary zoom with adequate contrast, without relying only on color, an icon, or image metadata, and include it in keyboard and screen-reader reading order.
• Preserve the meaning of all three elements across locale switching, and do not expose an image in a locale whose disclosure is absent or materially weaker.

In the mobile reading order examined by the assessment, the images appeared before the common disclosure. A later shared notice was not sufficient when a human-like image was encountered first, so the downstream UI constraint required either moving the full disclosure before the images or adding a visible AI-persona and AI-generated label to every image that could appear first.

Every image addition, removal, replacement, or move required another check of the AI-generation fact and its relationship to the disclosure. Because the design showed nine profiles while the asset inventory contained ten files, the assessment also prohibited making a public profile-count claim until the final set was resolved.

The claim boundary excluded unsupported biographies, careers, achievements, quotations, personality, wording that implied human staff or a real person, unverified product information, availability, and link destinations. Calls to action, contact routes, and policy surfaces could not appear available until the accountable owner confirmed the facts and operating process.

The pre-publication questions covered each image's generation method and later edits, provenance, rights and permitted use, real-person similarity and any required consent, publication territories and reachable readers, operator information, every display context, and the final profile set. Where the applicable territory or publication risk required qualified professional judgment, the result could not be described as legally approved until that assessment was completed separately.

The primary diff was checked line by line to confirm that the behavior described above matched the change at the cutoff.