Google Signs EU Code on Labeling AI-Generated Content

Google said on July 24, 2026 that it will sign the European Union’s voluntary Code of Practice on Transparency of AI-Generated Content, the framework Brussels is offering companies to meet transparency rules that become binding across the bloc on August 2, 2026. The decision commits one of the world’s largest providers of generative AI to the EU’s preferred route for labeling synthetic media, even as the company warned the regime could end up confusing the people it is meant to protect.

A voluntary route to a binding rule

Signing the code is optional. The obligations behind it are not. From August 2, 2026, the EU AI Act requires providers of systems that generate synthetic audio, images, video or text to mark those outputs in a machine-readable format so they can be detected as artificially generated. Businesses that deploy such systems will have to clearly label deepfakes, and AI-generated text published to inform the public on matters of public interest, unless a human editor takes responsibility for the content.

The code was drawn up by independent experts convened by the EU’s AI Office and published on June 10, 2026. It divides those duties into two parts: one set of commitments for providers that mark content at its source, another for the businesses that deploy AI systems and label what reaches the public. The European Commission and the member states’ AI Board have already assessed it as an adequate way to comply, which means a signature carries legal weight rather than symbolism. The code does not mandate any single technology; it expects a mix of machine-readable metadata, watermarking and provenance tracking, and leaves the exact method to the provider.

For a company, signing buys predictability. Adherents get one recognized compliance path across all 27 member states, and the AI Office has said future enforcement for signatories will focus on monitoring whether they honor the code. Companies that decline are not breaking any law by staying out, but they take on the burden of proving to national regulators that their own marking and labeling measures are adequate, and can expect more requests for information. Enforcement falls mainly to national market surveillance authorities, and breaches of the transparency rules can draw fines of up to 15 million euros or 3% of a company’s worldwide annual turnover.

The window to lock in the initial benefits is narrow. The AI Office asked organizations to file their signature forms by July 22, 2026 to appear on the first public list of signatories before the rules apply. Separately, EU lawmakers have agreed a grandfathering rule that gives generative systems already on the market before August 2 until December 2, 2026 to meet the machine-readable marking requirement. The reach extends past Europe: a provider based anywhere is covered once its AI outputs are used in the EU.

Google’s tools, and its objections

Google cast the move as a continuation of its 2025 signing of the EU’s separate code for general-purpose AI models, and tied it to watermarking work it has backed for years. The company said it is speeding up adoption of C2PA, an industry standard for tracking content provenance, and pointed to SynthID, its own technology for watermarking AI-generated images, audio, video and text. It said it is working with Apple (AAPL ), ElevenLabs, Kakao, Nvidia (NVDA ) and OpenAI to make those tools interoperable across platforms, the sort of cross-vendor detection the EU rules assume but for which no single standard is settled.

The endorsement came with a caveat. Karen Massin, who leads Google’s public policy for EU institutions, wrote that adding regulatory complexity while the technology is still evolving could work against Europe’s own competitiveness and simplification goals. If online content ends up “flooded with overlapping AI labels and legal disclosures,” she wrote, people will find it harder, not easier, to judge what they are seeing.

That critique targets a live tension in Brussels. The Commission spent much of 2026 trying to thin its digital rulebook, and the grace period for existing systems grew out of that same simplification drive. Google’s stance, backing the principle while questioning the execution, tracks the broader industry argument that the transparency rules could become a compliance chore before they deliver real clarity for the public.

What comes next

Platforms have been edging toward user-facing disclosure on their own; TikTok rolled out user controls for AI-generated content in its feeds. Governments are assembling parallel machinery, with the UK naming a chair for a new government-wide AI taskforce, but the EU’s law is the first to attach hard marking and labeling duties, and penalties, to synthetic media.

For now, the code is the closest thing to a settled answer on how to satisfy those duties, and a signature from a provider of Google’s scale makes it harder for smaller developers and rival labs to treat the framework as optional. The sharper test comes after August 2, when regulators begin judging whether the labels people actually see on AI-generated content are clear enough to count.