The Science Talk Signs the EU Code of Practice on Transparency of AI-Generated Content

We signed the European Commission’s Code of Practice on Transparency of AI-generated Content, under Section 2, as a deployer of generative AI systems. That puts The Science Talk in the same signatory pool as Bulgari, Getty Images, Lenovo, and Lufthansa. Section 1 of the framework counts Anthropic, Google, Meta, Microsoft, OpenAI, and Mistral among its signatories.

Why this, why now

The AI Act’s transparency obligations become legally binding on 2 August 2026. These are the rules that require AI-generated and AI-modified content to be marked and disclosed. The Commission approved the Code of Practice as the sanctioned route to compliance. Independent experts drafted it. The Commission and the AI Board found it sufficient. It gives signatories “a streamlined, predictable and legally certain pathway,” regardless of which national authority ends up supervising them.

By the Commission’s own count, about 190 organizations signed as of end of July 2026. We are one of them. About half of that signatory base is small, recently founded companies. This is not a club reserved for platforms with in-house legal teams. It is a framework built to scale down as well as up.

What Section 1 and Section 2 actually mean

The Code splits obligations by role, not by size:

Section 1: Providers. This covers providers of generative AI systems, the companies that build tools to mark and detect AI content, and AI model providers themselves. This is the technical backbone: the commitments that make labeling and detection possible at the infrastructure level. Anthropic, Google, Meta, Microsoft, OpenAI, and Mistral sit here. These are the organizations whose models and systems generate the content in the first place.

Section 2: Deployers. This covers organizations that put generative AI systems to work in production. Deployers use AI tools to create, adapt, or publish content for real audiences. Deployers must mark AI-generated content in the correct way. Deployers must disclose its use where it matters. Deployers must build practices that support public trust and do not erode it. This is where The Science Talk signed.

The distinction matters for how we frame this: we are not a lab that builds AI models. We are an agency that uses AI systems in science communication work. Section 2 is the commitment that we do that transparently, every time, not just when convenient.

What it says about how we operate

Science communication runs on trust. An audience can lose trust the moment it suspects an AI-generated graphic, script, or summary was shown as fully human-made without disclosure. That trust does not return with an apology. The Science Talk signed this Code before the legal deadline. This is a public commitment that The Science Talk treats trust as non-negotiable. The Science Talk does not treat it as a rule to obey only when the law requires it.

This is also a signal about pace. The legal obligation lands 2 August 2026. The Commission’s implementation task forces do not even launch until September 2026. We signed in the window before either of those dates, while most organizations in our space still must decide whether this applies to them.

Our disclosure practices are set out on our AI disclosure page. It states how The Science Talk marks and discloses AI use in its work.

What’s next

Task forces to support Code implementation launch in September 2026. We will track that process. We will update our own disclosure practices as the framework matures. We have tracked the AI Act’s high-risk timeline shifts and Article 4 literacy obligations the same way elsewhere on this site.

For now: if you work with The Science Talk, you can expect AI-generated content to be marked as such, consistently and without you having to ask.