Loading
AI transparency is the principle that people can know when an AI system is involved and get meaningful information about it. NIST's AI Risk Management Framework treats transparency as a distinct problem from explainability (describing how a system works) and interpretability (understanding what its output means), a system can be clearly disclosed as AI while still being a technical black box. The EU AI Act mandates specific transparency duties from 2 August 2026 under Article 50: chatbot disclosure, deepfake labelling, and machine-detectable AI-generated content.
Run the free AI Health CheckAI Transparency, the legal and ethical requirement that people are told when they are interacting with AI, when content is AI-generated, and, in some cases, how an AI decision affecting them was reached.
AI transparency obligations are crystallising fast. EU AI Act Article 50 (effective 2 August 2026) requires disclosure when users interact with AI chatbots, when content is deepfake or AI-generated synthetic content, and when emotion recognition or biometric categorisation is used. GDPR Article 22 already gives data subjects rights to explanation in automated decision-making contexts. Australia's Privacy Act ADM transparency obligations come into effect on 10 December 2026.
Source: EU AI Act, Article 50; GDPR, Article 22; Australian Privacy Act
These terms are frequently used interchangeably, but NIST's AI RMF (and the OECD AI Principles) treat them as distinct. Conflating them is a common source of confusion in both compliance programs and vendor marketing.
Transparency
Does the person know an AI system is involved?
NIST defines it as the extent to which information about an AI system and its outputs is available to the people interacting with it. It is about disclosure, not mechanism, a system can be transparent (clearly labelled as AI) while still being a technical black box.
Explainability
Can we describe the mechanism?
A representation of the internal mechanisms underlying how the system operates, what drove a specific output, and how the parts fit together.
Interpretability
Can we make sense of what the output means?
The meaning of the system's output in the context of its designed purpose, distinct from the mechanism itself; a result can be interpretable to a domain expert without being fully explainable to a lay user.
Article 50 applies regardless of a system's risk tier and is due from 2 August 2026, a date the Digital Omnibus (Council final approval 29 June 2026) did not postpone, unlike the separate high-risk Annex III obligations it pushed to December 2027.
Chatbots and conversational AI must disclose that a person is interacting with an AI system, unless this is obvious given the context.
Generative AI outputs (synthetic audio, image, video, text) must be marked in a machine-readable format, detectable as artificially generated. A narrow grace period to 2 December 2026 applies to systems already on the market before 2 August 2026.
Deployers of emotion-recognition or biometric-categorisation systems must inform the people exposed to them, and handle any personal data under GDPR.
Deepfake image, audio, or video content must be disclosed as artificially generated or manipulated, with a lighter-touch exemption for evidently artistic, satirical, or fictional work.
AI-generated text published to inform the public on matters of public interest must be disclosed, unless it underwent human editorial review with a named person or organisation holding editorial responsibility.
Chatbot disclosure, deepfake labelling, machine-readable content marking, public-interest text disclosure. Not postponed by the Digital Omnibus (which affects only the separate high-risk Annex III tier).
Requires both explicit labels (visible/audible markers users can perceive) and implicit labels (technical metadata markers) on AI-generated text, images, audio, video, and virtual scenes.
Covered providers (over 1 million monthly California users) must offer a visible "manifest disclosure" label and embed hidden "latent disclosure" metadata in AI-generated content, plus a free public detection tool. Date deliberately aligned with the EU AI Act.
What is AI transparency?
AI transparency is the principle that people should be able to know when they are interacting with, or subject to, an AI system, and to access meaningful information about how it works and was built. NIST's AI Risk Management Framework defines it as the extent to which information about an AI system and its outputs is available to individuals interacting with it.
What is the difference between AI transparency, explainability, and interpretability?
Transparency asks whether the person knows an AI system is involved and can get basic information about it. Explainability asks whether the underlying mechanism can be described. Interpretability asks whether the output's meaning can be understood in context. NIST treats "Accountable and Transparent" and "Explainable and Interpretable" as two separate trustworthy-AI characteristics precisely because disclosure is a distinct problem from mechanistic understanding: a system can be labelled as AI (transparent) while remaining a technical black box (not explainable).
What does the EU AI Act require for AI transparency?
Article 50 imposes disclosure duties regardless of a system's risk tier: chatbot providers must ensure people know they are talking to AI; generative AI providers must mark synthetic output in a machine-readable format; deployers of emotion-recognition or biometric-categorisation systems must inform exposed individuals; and deepfake or AI-generated public-interest text must be disclosed, subject to artistic-work and editorial-review exemptions.
Were the EU AI Act transparency rules delayed by the Digital Omnibus?
No. The Digital Omnibus (Council final approval 29 June 2026) postponed the separate high-risk (Annex III) compliance obligations to December 2027/August 2028. It left Article 50's 2 August 2026 date essentially untouched, with only a narrow grace period to 2 December 2026 for the content-marking duty on generative AI systems already on the market before 2 August 2026.
Does the US have federal AI transparency rules?
No single federal statute. California's AI Transparency Act (SB 942, amended by AB 853) is the most developed state law, requiring large generative-AI providers to offer visible and hidden AI-content disclosures. Its operative date was moved to 2 August 2026, deliberately timed to align with the EU AI Act.
How does China regulate AI content transparency?
China's Cyberspace Administration, with three other ministries, issued Measures for the Labelling of AI-Generated Synthetic Content alongside the mandatory technical standard GB 45438-2025, both effective 1 September 2025. They require both visible/audible labels and embedded technical metadata markers, a dual model conceptually similar to the EU and California approaches.
Last reviewed July 2026