EU starts mandatory AI generated content watermarking from August 2nd

The EU’s new AI Act comes into force on August 2nd, 2026 and will mandate that AI generated content that looks real has a digital watermark. This includes images, sounds and text, but excludes personal messages and obviously artistic/satirical content.

Non-compliance could result in fines of up to 3% of a company’s worldwide annual revenue. If you have an AI system running in the EU then you get another four months to comply. You can use the suggested labelling system proposed by the EU, or develop your own as long as you meet the technical requirements.

It’s all about protecting consumers and avoiding misinformation. According to the MEP who pushed this legislation through, it’s “a weapon to fight disinformation and preserve democracy and the truth online”. It’s mostly about synthetic media, which without a watermark will be indistinguishable from real content like deepfakes, synthetic voices and AI written stories.

This is the first significant law addressing deepfakes and misinformation and having real world impact. Global tech firms like Google and TikTok now need to make sure they have systems in place to implement watermarks. Other organisations will no doubt follow suit (the UK and Canada are watching closely).