Tech companies must label AI content from Sunday

Video, imagery, audio and even text must be marked

The EU is introducing new rules requiring companies to clearly identify content created or significantly altered using AI.

The measures, which take effect on Sunday, form part of the bloc's landmark AI Act and apply to companies operating in or targeting the EU market.

They require providers to disclose when images, audio, video or even text have been generated or substantially modified by AI, using visible labels alongside technical methods such as digital watermarks and embedded metadata.

The rules are intended to help internet users distinguish between authentic and AI-generated material, as these increasingly sophisticated tools make it more and more difficult to tell the difference.

Platforms already adapting

Several of the world's largest technology companies have already introduced measures that broadly align with the new requirements.

Meta has rolled out "AI Info" labels on Facebook and Instagram to identify posts created with generative AI.

TikTok also now requires creators to disclose AI-generated material, saying its detection tools have tagged billions of AI-generated posts.

Google has signed the European Commission's voluntary code of practice on AI transparency and is working with industry partners, including Nvidia, Apple and OpenAI, on standards designed to preserve information about the origin of digital content.

Under the new rules companies are expected to do more than simply label AI content. They also have to implement technical systems that can identify AI-generated material - even after it has been shared or reposted across different platforms.

Brussels has stressed that the legislation is not intended to cover every use of synthetic media.

AI-generated content produced for clearly artistic, fictional, satirical or other creative purposes will generally fall outside the disclosure requirements.

Tackling the risks of deepfakes

There have been a series of high-profile incidents showing the technology's potential for abuse in recent years.

In 2024, explicit AI-generated images falsely depicting pop star Taylor Swift circulated widely online without her consent.

That same year, a staff member at Arup fell for a scam that cost the engineering giant £20 million.

And in the opening weeks of Russia's full-scale invasion of Ukraine in 2022 – before ChatGPT even came onto the scene - a fake video appeared to show President Volodymyr Zelensky urging Ukrainian troops to surrender.

Although it was quickly exposed as fake, the incident illustrated how synthetic media could be deployed during armed conflict to spread disinformation.

Technical obstacles remain

Despite broad support for AI transparency, technical and legal challenges stand in the way of the rules.

One such challenge is jurisdiction. While the EU can regulate companies serving European users, AI-generated content created elsewhere can rapidly spread across borders through global platforms.

There is also the issue of fragmentation. No universal standard exists yet for marking AI-generated content, and companies are experimenting with different approaches including watermarking, metadata and content provenance systems. Many of these are not fully compatible with one another.

Researchers have also warned that digital markers can be stripped away or degraded when files are edited, compressed or repeatedly shared, making it difficult to trace synthetic media’s origins.

A technical study commissioned by the European Commission concluded that combining multiple methods of detection and marking is likely to be more effective than relying on any single technology.

Some tech firms also have concerns about the growing complexity of AI regulation.

Google executive Karen Massin has warned that an increasing number of labels and disclosures could ultimately confuse users rather than improve transparency if they become too commonplace.

Companies have until 2nd December to ensure existing AI systems comply with the new obligations. After that, regulators will be able to impose financial penalties on organisations that fail to meet the requirements.