Cohere has become one of the first signatories to the European Union's Code of Practice on transparent AI-generated content. The company signed Section 1 of the voluntary code, which applies to providers of AI systems and is intended to help them prepare for transparency obligations under Article 50 of the EU AI Act.
The commitment concerns how providers make AI-generated or AI-manipulated content identifiable. It is a governance announcement rather than a new model or product release, and Cohere has not announced pricing, availability or feature changes as part of the decision.
What Cohere has committed to
Cohere says it will follow transparent practices and support appropriate marking of AI-generated content. The code is intended to give providers a practical route towards meeting legal transparency requirements, while giving regulators, users and downstream organisations a clearer basis for understanding when content has been produced or altered by AI.
The company signed only the section applicable to AI system providers. That scope matters because different participants in the AI supply chain may have different responsibilities. A model developer, an application provider and a user publishing synthetic media do not necessarily control the same technical or distribution points.
Cohere's move follows its earlier signature of the EU's general-purpose AI Code of Practice. The company also says it is engaging with the EU AI Office through an Advisory Forum and a Signatory Taskforce. This continuing involvement should give Cohere a role in practical discussions about how transparency measures are interpreted and implemented.
Why content transparency matters
AI-generated text, images, audio and video can be useful in business and creative workflows, but they can also make it harder to judge provenance. Marking mechanisms can help platforms, organisations and audiences recognise synthetic content, investigate misuse and apply their own policies.
No single technique is sufficient for every format or distribution channel. Visible labels can be removed, metadata can be stripped and technical watermarks may degrade when content is compressed or edited. Effective compliance is therefore likely to require several layers, including machine-readable signals, durable provenance records, user-facing disclosures and documented handling by downstream applications.
Enterprise customers will need to understand which layer Cohere controls and which obligations remain with them. An organisation that uses a model to generate customer communications, marketing assets or public information may still need policies for review, disclosure, record keeping and distribution even if the underlying provider supports technical marking.
What remains to be clarified
Cohere's announcement does not describe a technical rollout, implementation timetable or the exact mechanisms it will use across its models and enterprise products. It also does not establish how markings will persist when outputs are copied into another tool, transformed by a separate model or exported into common file formats.
The code is voluntary, but it is designed around binding EU AI Act requirements. Signing indicates intent and creates a public accountability point; it does not by itself demonstrate compliance in every deployment. The practical value will depend on the resulting product controls, documentation, interoperability and evidence available to customers and regulators.
For organisations procuring Cohere services in Europe, the announcement is a reason to ask more specific implementation questions. Buyers should request details about supported content types, provenance metadata, disclosure interfaces, retention, audit evidence and the allocation of responsibilities between Cohere and the deploying organisation.
Cohere's early signature adds momentum to an emerging common approach for AI-content transparency. The next meaningful milestone will be technical documentation showing how the commitment operates in real products and how reliably the signals survive ordinary publishing and editing workflows.