Using AI? These Things Are Banned Starting August

Starting August 2, 2026, the European Union’s AI Act introduces strict transparency rules, banning concealed chatbot interactions, unmarked deepfakes, and unacknowledged automated public information materials.

New Transparency Rules Under the AI Act

From August 2, 2026, hiding that a recipient is talking to a machine, publishing unmarked professional deepfakes, and presenting automatically generated material as independent human work will no longer be permitted. The new phase of the AI Act focuses primarily on transparency.

The EU regulation does not introduce a general ban on text, image, or sound generators, nor does it prohibit companies from using chatbots. However, operating artificial intelligence invisibly to clients, employees, readers, or viewers will become significantly harder, governed by Article 50 of Regulation 2024/1689.

Restrictions on Chatbots and Direct Interaction

The first obligation applies to systems designed for direct human interaction, such as store chatbots, automated bank consultants, and voice assistants. Providers must design systems to inform users they are communicating with AI unless it is obvious to an average, reasonably informed person.

The European Commission clarifies that this obligation applies to real, two-way direct communication. It does not apply to simple forms or software that only collects data without conducting a conversation, meaning bots cannot be designed to mimic human staff silently.

Technical Tracking and Content Labeling

The second obligation targets providers of text, image, audio, and video generators. Outputs must be marked in a machine-readable format detectable as generated or modified by AI, utilizing metadata, cryptographic mechanisms, or digital proofs of origin.

Standard editing that does not substantially alter input data—such as spellchecks, noise reduction, or brightness adjustments—is exempt. Systems introduced before August 2, 2026, have a transitional period until December 2, 2026, to implement these technical markings.

Rules for Deepfakes and Professional Content

Distinct rules govern deepfakes, defined as manipulated or generated audio, video, or imagery resembling real entities. Businesses, media, freelancers, and influencers using such material professionally must clearly disclose its artificial origin at first exposure without relying solely on metadata.

Artistic, satirical, and fictional works receive a more lenient execution method where AI disclosure must not disrupt the work’s reception. Furthermore, public information texts generated or manipulated by AI require labeling unless they undergo genuine human or editorial control.

Biometrics, Prohibited Practices, and Penalties

Permitted emotion recognition and biometric categorization systems must inform subjected individuals, though workplace emotion recognition has been broadly banned since February 2025. An additional ninth ban targeting unauthorized sexual and intimate content generation tools will apply starting December 2, 2026.

Violations of transparency obligations can incur administrative fines of up to 15 million euros or 3 percent of total annual global turnover. Prohibited practices under Article 5 carry fines up to 35 million euros or 7 percent of global turnover.