EU publishes code for marking AI-generated content

New EU transparency rules will require developers and users to disclose when AI is used to create or alter images, video, audio and some text. A voluntary code sets out marking, metadata and labeling practices businesses can use to show compliance.

The European Commission has published a Code of Practice on Transparency of AI-Generated Content to help companies prepare for EU AI Act transparency duties that take effect on August 2, 2026. The rules will apply to many developers and users of generative AI systems, requiring disclosure when AI is used to create images, videos, audio and, in certain cases, text.

The code is organized around providers and deployers of AI systems. It recommends durable marking methods, including hidden watermarks embedded in content and cryptographically signed digital metadata. Providers are expected to preserve existing metadata, avoid altering or removing tags, prohibit tampering in acceptable use policies and make markings resilient to typical processing operations.

The guidance also discourages providers from offering, promoting or advertising tools designed to remove or bypass content markings. For deployers, it recommends consistent labeling for deep fakes and AI-generated text, including clear disclosures about what was altered, visible icons or labels, and audio warnings when visual labels cannot be displayed.

Compliance with the code is voluntary, but the transparency requirements in Article 50 of the EU AI Act are legal obligations. Providers and deployers that sign the code may use its measures to demonstrate compliance once it receives a positive assessment, while organizations using other methods will need to show those measures are adequate.

72

Impact Score

SpaceX agrees $60bn deal to buy Cursor

The all-stock acquisition would bring Anysphere’s AI coding platform into SpaceX as Elon Musk’s companies deepen their enterprise AI push. Cursor counts Stripe, Adobe and Nvidia among reported corporate users.

ASML warns Terafab hinges on EUV supply

Christophe Fouquet has held direct talks with Elon Musk about Terafab, but ASML’s bigger concern is whether it can meet another wave of lithography demand. The proposed Texas fab could intensify competition for the machines behind advanced chip production.

Epic opens Unreal Engine to competing AI models

An experimental MCP plugin in Unreal Engine 5.8 gives LLMs a path into real project data and engine systems. The move sets up a workflow contest among model providers rather than a single default winner.

Researchers use model signals to detect AI hallucinations

New methods analyze activations, attention patterns and output probabilities to flag hallucinations, memorized data and unreliable responses. The work points toward monitoring systems that can catch failures as models generate text.

Contact Us

Got questions? Use the form to contact us.

Contact Form

Clicking next sends a verification code to your email. After verifying, you can enter your message.