From 2 August, companies that build or use AI systems in the European Union must label content that looks authentic but was machine-made, as the AI Act’s transparency rules move into enforcement. The rules apply to providers and deployers across the bloc and cover deepfakes, synthetic media, machine-generated text, and chatbot interactions.

Providers, meaning firms that build generative systems, must embed machine-readable markings so images, audio, video, and text can be detected as AI-generated or manipulated. Deployers, meaning people or companies using AI systems, must clearly and promptly disclose when they create a deepfake or publish machine-written text on matters of public interest.

The duty to label realistic deepfakes applies regardless of whether the creator intended to deceive. People must also be told at the moment they start interacting with a chatbot, rather than through terms-of-service language. Clearly unrealistic or fantastical content is outside the rules, and AI-generated text avoids labelling if a human with genuine editorial responsibility has meaningfully reviewed it.

Artistic and satirical content still requires disclosure, but the notice must not spoil the experience of the work. Breaches can bring fines of up to €15m or 3% of a company’s worldwide annual turnover, whichever is larger. The obligations formally apply from 2 August, though a simplification package has floated moving the machine-marking deadline to December.

Advertisers have argued that AI-generated ads should be exempt from the transparency rules. The European Commission published draft guidelines in May and is finalising a code of practice for marking and labelling AI content, which companies can sign to show compliance.

The rules may affect platforms beyond Europe because global services rarely build separate systems for one market. Enforcement remains difficult because watermarks can be stripped, metadata can be lost, and machine-written text is difficult to detect. The EU has separately moved to ban non-consensual intimate deepfakes, while the wider labelling regime targets synthetic media that can undermine trust online.

From 2 August, companies that build or use AI systems in the European Union must label content that looks authentic but was machine-made, as the AI Act’s transparency rules move into enforcement. The rules apply to providers and deployers across the bloc and cover deepfakes, synthetic media, machine-generated text, and chatbot interactions.

Providers, meaning firms that build generative systems, must embed machine-readable markings so images, audio, video, and text can be detected as AI-generated or manipulated. Deployers, meaning people or companies using AI systems, must clearly and promptly disclose when they create a deepfake or publish machine-written text on matters of public interest.

The duty to label realistic deepfakes applies regardless of whether the creator intended to deceive. People must also be told at the moment they start interacting with a chatbot, rather than through terms-of-service language. Clearly unrealistic or fantastical content is outside the rules, and AI-generated text avoids labelling if a human with genuine editorial responsibility has meaningfully reviewed it.

Artistic and satirical content still requires disclosure, but the notice must not spoil the experience of the work. Breaches can bring fines of up to €15m or 3% of a company’s worldwide annual turnover, whichever is larger. The obligations formally apply from 2 August, though a simplification package has floated moving the machine-marking deadline to December.

Advertisers have argued that AI-generated ads should be exempt from the transparency rules. The European Commission published draft guidelines in May and is finalising a code of practice for marking and labelling AI content, which companies can sign to show compliance.

The rules may affect platforms beyond Europe because global services rarely build separate systems for one market. Enforcement remains difficult because watermarks can be stripped, metadata can be lost, and machine-written text is difficult to detect. The EU has separately moved to ban non-consensual intimate deepfakes, while the wider labelling regime targets synthetic media that can undermine trust online.