Starting August 2, 2026, the next phase of the EU’s AI Act kicks in, and it targets one of the internet’s most unsettling problems: synthetic media that pretends to be real. Any AI-generated content impersonating actual people or real events will need to be clearly labelled, and the companies behind it — social media platforms very much included — either comply or face serious penalties.
The mechanics are refreshingly flexible. According to Le Monde, there’s no single mandated method: watermarks, visible labels, embedded metadata — anything goes, provided the result makes synthetic content easy to detect. The EU’s real concern is narrower than it might sound. It’s not chasing every AI meme; it’s after deepfakes engineered to imitate reality and deceive people at scale.
“Generative AI enables disinformation to be created at unprecedented scale, tailored to specific audiences, and disseminated with remarkable speed,” an EU official told Le Monde. The stated goal is to help citizens keep trusting what they see, hear and read.
Crucially, the rules apply to content made for professional purposes, including AI-generated informational text meant for public consumption that had no human oversight. If you’re producing AI content for artistic, satirical or fictional reasons, or purely for private use, you fall outside the law’s scope. Nobody’s coming for your surreal AI art project.
The fines, however, are not trivial. Breaches can cost up to €15 million or 3% of global annual turnover — the kind of number that concentrates a boardroom’s attention.
Big Tech’s response has been a study in cautious cooperation. Google has signed the EU AI Act code of conduct on AI transparency, while Meta has signalled its intention to sign. Both are hedging. Google supports transparency but frets that “adding more regulatory complexity — as technical solutions are still evolving — could contradict Europe’s goals for competitiveness and simplification.” Meta warns against burying users under “a growing array of different labels and disclosures” that overwhelm people while piling regulatory weight onto providers.
The platforms aren’t starting from scratch, though. Meta rolled out an “AI Info” label in 2024, and TikTok has required creators to flag AI imagery for several years. The tools are largely in place.
The harder questions are technical and jurisdictional. Someone outside the EU can generate a deepfake that European citizens see within minutes — and in that scenario, the responsibility shifts to the platform. If its detection tools miss the fake, the law’s protection quietly evaporates.
And the underlying tech is a moving target. The industry currently leans on the C2PA protocol and Google’s invisible SynthID watermark, but every one of these systems has known weaknesses. What reliably flags synthetic content today may be trivially defeated a few months from now. The EU has drawn a line in the sand; whether the technology can hold it is another matter entirely.
