European Commission sets August 2 AI transparency deadline
TL;DR
- The Commission's July 20 FAQ confirms chatbot disclosure must occur at the moment of first contact, not through terms of service or privacy policy references.
- The deepfake labeling obligation applies regardless of deceptive intent, binding marketing teams and content platforms alongside bad actors.
- Companies signing the Code of Practice by July 22 gain presumption of regulatory conformity, shifting the burden of proof to enforcement authorities rather than the company.
European Commission executive vice-president Henna Virkkunen has put a date on when the AI Act's transparency rules stop being aspirational and start biting: August 2. The Register reports the Commission published guidelines on July 20 walking providers through what will be required, namely telling users when they are talking to a chatbot or an AI agent, adding machine-readable marks to AI-generated or manipulated audio, images, video and text, disclosing deepfakes to the people exposed to them, and labelling AI-generated text on matters of public interest unless a human editor reviewed it.
The sweep of the deepfake rule is the part worth reading twice. Content that looks or sounds like a real person has to be labelled even if no deception was intended and even if the person depicted is not real. That is a stiffer standard than 'do not mislead', and it puts a compliance floor under every synthetic-media pipeline in Europe rather than only the ones with bad intent.
Why this matters if you ship a product into the EU: August 2 is a hard date, and standard editing tools like spelling and grammar correction are exempt unless they substantially alter the input. That will comfort office-suite vendors, but anyone building a customer-facing bot, an image generator, or a synthetic-voice tool has a disclosure UX and a watermarking pipeline to ship in about two weeks. Virkkunen's line is that the guidelines are meant to make chatbots, AI agents and AI content 'more transparent and trustworthy', which reads as a signal that the Commission expects providers to treat this as a shipping deadline rather than a discussion draft.
The honest caveat is that the published document is a compliance roadmap rather than the regulation itself, and the reporting does not spell out fines, mandated watermarking standards, or which national regulators lead enforcement. It also does not resolve the 'editorially reviewed' carve-out for AI-generated text on public-interest matters, which is exactly the ambiguity publishers and platforms will spend the next year testing, alongside the postponed high-risk deadlines that now sit at December 2, 2027 for standalone systems and August 2, 2028 for embedded ones.
For provenance-tooling vendors selling watermarking and for publishers with real human editorial workflows, the direction is friendly. For everyone else building consumer-facing AI in Europe, the roadmap just got a lot shorter.
What others are reporting
-
European Commission Read →
The primary source document: the Commission's own official FAQ published July 20, confirming the disclosure obligation and its scope in the Commission's own words.
Providers of AI systems that directly interact with people must design and develop those systems in such a way that the individuals concerned are informed that they are interacting with an AI system, unless this is obvious.
-
European Commission Read →
The Commission's companion Code of Practice page, updated July 20, covers the voluntary compliance tool and the July 22 signing deadline that shifts the burden of proof to enforcement authorities.
These transparency obligations, applicable from 2 August 2026, complement other rules like those for high-risk AI systems.
-
Greenberg Traurig LLP Read →
Major US law firm analysis framing the disclosure obligation as a contact-time duty, not a fine-print exercise, confirming enforcement scope is wider than most product teams planned.
Users need to know they are interacting with AI at the moment of contact – not through a reference buried in fine print.
-
Global Policy Watch (Covington & Burling) Read →
Covington's structured breakdown adds that technical feasibility is judged as an objective standard, not relative to a given company's size or resources, closing a common compliance escape route.
The Guidelines clarify that technical feasibility is an 'objective notion that is not dependent on the specific resources and capabilities of individual providers.'
-
Bratby Law Read →
The only source providing a direct EU/UK comparative; UK businesses face parallel but distinct disclosure duties under GDPR, Online Safety Act, and FCA rules that satisfying Article 50 does not discharge.
The AI Act transparency obligations ask one thing of any business putting AI in front of people in the EU: say so.
-
Pebblous Read →
Adds that 78% of affected organizations had not started compliance work as of late June, and explicitly flags that the AI Omnibus delay applies only to high-risk AI, not Article 50.
A company that spends these six weeks well can get ahead in the procurement race for the EU market.
-
Zyphe Read →
Frames Article 50 as a transparency regime, not an anti-fraud control, and notes that criminal deepfake producers will not label their content, exposing the structural ceiling of the disclosure approach.
Article 50 is a transparency regime, not an anti-fraud control — and criminal deepfake authors won't label their content.
Originally reported by theregister.com
Read the original article →Original headline: European Commission Publishes AI Act Transparency Guidelines Ahead of August 2 Enforcement Deadline for Chatbots and Deepfakes