AI content disclosure - FAQ

EU AI Act Article 50 sets transparency duties that apply from 2 August 2026. People must be told when they are interacting with an AI system (for example a chatbot) unless it is obvious, and AI-generated or manipulated images, audio and video must be marked in a machine-readable way as artificially generated or manipulated. Deployers of deep fakes must also disclose that the content is artificially generated or manipulated. See the official text at eur-lex.europa.eu (Regulation (EU) 2024/1689).

Increasingly, yes. Under the EU AI Act, providers of generative AI must mark synthetic images in a machine-readable format, and in several jurisdictions visible disclosure is expected for manipulated media. The practical approach is two layers: a visible caption or badge (for example 'AI-generated image'), plus machine-readable marking (C2PA Content Credentials and the IPTC digitalSourceType field) so the label survives when the image is downloaded.

If users interact with an AI-powered chat and it is not obvious they are talking to a machine, you should disclose it. EU AI Act Article 50(1) requires informing people they are dealing with an AI system unless it is obvious from the context. California's B.O.T. Act (SB 1001) also requires disclosing bots used to incentivise a commercial transaction. A short line such as 'You're chatting with an AI assistant' at the start of the conversation is the usual fix.

California AB 723 requires real-estate agents (and people acting on their behalf) to clearly disclose when a marketing image of a property has been digitally altered - for example virtually staged or AI-enhanced - and, for images posted online, to provide access to the unaltered original via a link, URL or QR code. It applies from 1 January 2026 and distinguishes standard edits (like exposure) from alterations that change the representation of the property.

C2PA (Coalition for Content Provenance and Authenticity) is an open standard for attaching tamper-evident 'Content Credentials' to media, recording how it was created or edited - including whether AI was involved. Because the credential is embedded and cryptographically signed, the AI label can travel with the file when it is downloaded or shared, which is what several transparency rules ask for.

digitalSourceType is a standard IPTC metadata field that records how media was produced. Values such as 'trainedAlgorithmicMedia' mark content created by a generative AI model, and 'compositeWithTrainedAlgorithmicMedia' marks AI-edited media. Detectors from major platforms read this field, so writing it into an image is a simple, deterministic way to make an AI label machine-readable.

Metadata alone is fragile — it can be stripped by re-encoding or screenshots. The resilient approach is to combine layers, and Discloze's metadata embedder applies up to three: IPTC digitalSourceType, C2PA Content Credentials (cryptographically signed provenance), and an invisible watermark embedded in the pixels. The watermark survives recompression and re-encoding (though not resizing or cropping), so even if metadata is lost, the AI marker can still be detected.

Yes — Discloze's free Content Credentials verifier lets you upload an image or paste its URL and instantly see whether it carries C2PA Content Credentials, IPTC/EXIF AI metadata, or an invisible watermark. Note that screenshots and re-encoding can strip C2PA and IPTC, so check the original file; the absence of a marker does not prove content is not AI-generated.

Discloze crawls a site (sitemap-first, respecting robots.txt) and checks each image with layered detection, most reliable first: C2PA manifests, IPTC/EXIF metadata, then weaker signals like filename or alt-text hints. It also detects chat widgets and checks whether an AI disclosure is present. Deterministic metadata signals are reported as detected; weaker signals are only ever shown as 'likely AI', never as certain.

No. Discloze is compliance tooling, not legal advice, and cannot guarantee compliance with any law. Automated detection is best-effort and may miss or misclassify content, and rule summaries are for guidance only. It helps you find and fix likely gaps and keep an audit trail - but you should confirm important decisions with a qualified professional.

The rule packs currently cover 15 jurisdictions: the EU (AI Act Article 50), Italy (Law 132/2025), Spain (draft), the UK (advisory), Australia (advisory), the US states of California, New York, Pennsylvania, Washington, Colorado, Utah and Texas, China, and South Korea. Each rule cites its source and effective date; rules that are draft, advisory, or not yet in force are shown for information without affecting the score (and in-force dates activate automatically). New jurisdictions are added as data, without code changes.