Google announced on July 24, 2026 that it has formally signed the European Union’s AI Act Code of Practice on Transparency of AI-Generated Content, committing the company to a set of technical and labelling standards that will govern how AI-produced text, images, audio and video are identified across Europe.
At the same time, Google disclosed that it is partnering with third-party AI labs including Apple, ElevenLabs, Kakao, NVIDIA, and OpenAI to drive industry-wide adoption of interoperable watermarking tools using SynthID.
What the code requires
The Code of Practice on Transparency of AI-Generated Content helps providers and deployers of generative AI systems comply with the AI Act’s obligations for labelling and marking of AI-generated content.
Those obligations, set out in Article 50 of the EU AI Act, cover both marking AI-generated or manipulated output so it is machine-detectable and clearly labelling deepfakes and AI-generated text publications on matters of public interest.
On July 8, the European Commission concluded, by way of an opinion, that the Code of Practice adequately covers the obligations provided for in Articles 50(2), (4) and (5) of the AI Act and facilitates their effective implementation. Article 50 will become applicable on August 2, 2026, giving signatories a narrow window to align their systems.
The code was developed by independent experts in a multi-stakeholder process coordinated by the AI Office, involving participants from industry, academia, civil society, rightsholders, and EU Member States.
Adherence to the code is voluntary, but it provides an EU-wide recognised practical framework for signatories. Providers and deployers who choose not to sign will need to demonstrate compliance through other means and may face greater scrutiny.
SynthID goes cross-platform
Google is signing the code to help support the responsible use of AI in Europe, building on its 2025 signing of the GPAI Code of Practice and aligning with ongoing work to adopt and accelerate the C2PA industry standard and to develop AI transparency tools, including SynthID.
SynthID is Google’s industry-leading digital watermarking technology that helps people understand how content was created and edited. The technology works by embedding signals into AI-generated content that are invisible to humans but detectable by verification tools, and these signals survive common editing operations like cropping, compression, and color adjustment.
Google has open-sourced its SynthID text watermarking technology, and the primary SynthID open-source repository lives under the Google DeepMind GitHub organization, containing Python implementations of the core text watermarking algorithms, evaluation scripts, and dataset utilities.
To promote trust across platforms, Google has been partnering with third-party AI labs including Apple, ElevenLabs, Kakao, NVIDIA, and OpenAI to drive industry-wide adoption of interoperable watermarking tools using SynthID.
NVIDIA had already begun deploying SynthID in 2025, making the GPU giant an early adopter. The addition of Kakao, South Korea’s dominant messaging and content platform, extends the technology’s reach into Asian markets. OpenAI is integrating SynthID watermarking into images generated through ChatGPT, Codex, and its API. ElevenLabs will use it for AI-generated audio and voice. That breadth of partners spans enterprise AI hardware, creative audio generation, a major South Korean messaging platform, and leading large-language-model providers, making SynthID one of the most cross-industry watermarking efforts yet attempted.
The move also connects directly to the broader challenge of synthetic data scarcity that shapes the AI industry. As researchers warn that the internet is running low on genuine human-written training material, clear machine-readable provenance signals become more important for distinguishing original human content from AI re-generations. Watermarking tools like SynthID are increasingly positioned as the technical layer that makes that distinction reliable at scale.
C2PA as the open standard backbone
Alongside SynthID, Google is backing the Coalition for Content Provenance and Authenticity (C2PA) standard, an open technical specification developed collaboratively by Adobe, Microsoft, the BBC and others that embeds cryptographic provenance metadata directly into media files.
C2PA metadata can be stripped by a screenshot or re-encode, but it carries rich provenance context. SynthID survives screenshots and compression but carries almost no information on its own. Combining them means a file that loses its C2PA metadata can still be flagged as AI-generated through SynthID, while a file that retains its C2PA metadata offers full provenance detail.
The Code of Practice itself calls on signatories to implement a multilayered approach of active marking techniques that can be applied at different stages of the content-creation value chain, from model providers to third-party transparency specialists. Adopting both SynthID watermarking and C2PA metadata satisfies that multilayer requirement across different content types.
Developer implications
By expanding SynthID beyond its own Gemini models through open-sourcing the text watermarking component and partnering with outside developers, Google aims to help other developers integrate the watermarking technology within the output of their own generative AI systems.
The open text watermarking repository lowers the barrier for smaller studios and independent model builders who lack the resources to develop proprietary provenance tools.
For developers already building on Google’s ecosystem, the signing reinforces obligations that will be practically enforceable from August 2. Google’s Gemini model family already embeds SynthID across text and image outputs, giving partners a proven production reference implementation that third-party developers can study and extend.
The combination of a regulatory deadline, an expanding partner network and a widening coalition of adopters places SynthID at the center of what is fast becoming a mandatory layer of modern AI deployment in Europe.
