The voluntary code provides an approved way to meet binding EU requirements for marking AI output and labeling deepfakes and certain public-interest text.
The European Commission’s live signatory list now names 190 organizations that have backed its voluntary Code of Practice on Transparency of AI-generated Content.
The code gives AI providers and companies using generative AI an approved way to show how they meet European Union requirements for marking and labeling AI-generated content.
It divides responsibility between providers, which develop and offer AI systems, and deployers, which use those systems for business or professional purposes. Organizations can sign the section that applies to them or both.
AI providers must make outputs detectable
Providers must place machine-readable markers in AI-generated or manipulated text, images, audio, and video. These digital markers must allow automated tools to identify the material as AI-generated content.
The requirement applies to providers offering AI systems in the European Union. Providers based elsewhere are also covered when their systems’ output is used within the bloc.
The code gives signatories a common set of measures they can use to demonstrate that their marking and detection methods comply with the AI Act.
Companies publishing AI content face separate labels
Organizations using generative AI must clearly label deepfakes and certain AI-generated or manipulated text.
The deepfake requirement covers images, audio, and video that resemble real people, objects, places, organizations, or events and could appear authentic.
Companies must also label AI-generated text on matters of public interest, including politics, public health, finance, science, and culture. No label is required if the text receives substantive human review and someone accepts responsibility for publishing it.
A spelling or grammar check does not qualify as human review. The Commission also said a machine-readable marker added by the AI provider does not replace the clear label required from the company publishing the content.
Signatories span technology and other industries
The provider section currently has 82 signatories, while the section for companies using AI has 152. Some organizations signed both, leaving 190 distinct names on the Commission’s list.
Anthropic, Google, Meta, Microsoft, and OpenAI are among the companies backing the provider requirements. Bulgari, Getty Images, Lenovo, and Lufthansa are among those supporting the rules for organizations using and publishing AI-generated content.
Signatories also come from the information technology, telecommunications, education, and retail sectors. The Commission described about half of the signatories as small, recently established companies.
The code is voluntary, but EU requirements are not
Signing the code is voluntary. The underlying AI Act transparency requirements became legally binding on August 2.
Signatories can rely on the code to demonstrate compliance across EU member states. Organizations that do not sign must show compliance through other adequate measures and may receive more requests for information from regulators.
AI systems placed on the market before August 2 have until December 2 to meet the machine-readable marking requirement. The grace period does not apply to the other transparency requirements.
National regulators will handle most enforcement. Violations can bring fines of up to €15 million or 3% of worldwide revenue from the preceding financial year.
In September, the European Union’s AI Office plans to launch two task forces and invite signatories to participate. Members will compare how they are applying the transparency measures and explore better ways to put the code into practice.
