EU rules requiring labels on AI-generated content take effect

European Union rules requiring companies to identify AI-generated and manipulated material took effect Sunday as part of the bloc’s sweeping Artificial Intelligence Act.

Article 50 of the law requires providers of chatbots and other interactive AI systems to tell people when they are communicating with AI, unless that fact is already obvious.

Providers of systems capable of producing synthetic text, images, video or audio must also embed machine-readable markers that allow the material to be detected as AI-generated or manipulated, according to the European Commission.

Companies and other professional users deploying AI systems must provide clear and perceivable labels when publishing deepfakes or AI-generated text on matters of public interest that has not undergone meaningful human review or editorial control.

They must also notify people exposed to AI-based emotion recognition or biometric categorisation systems.

The commission said the requirements were intended to reduce deception, impersonation and misinformation by allowing people to determine whether the material they see, hear or read is authentic.

Clamp down on deepfakes

Deepfakes are defined under the law as AI-generated or manipulated images, audio or video resembling existing people, objects, places or events and falsely appearing to be authentic.

AI-generated text concerning politics, public health, security, the environment, consumer safety, science, culture and other subjects of public debate must be labelled when published without human editorial oversight.

Material that has undergone substantive human review does not require a label. The commission stressed, however, that superficial checks limited to spelling, grammar or formatting would not qualify as editorial control.

Special provisions apply to artistic, creative, satirical and fictional works containing deepfakes. Disclosure is still required but can be presented in a way that does not interfere with the enjoyment or display of the work.

People using AI for purely personal and non-professional purposes are outside the law’s scope. Individuals who regularly use it for economic gain or as part of a business, trade, freelance or professional activity may be considered deployers and therefore covered by the requirements.

The rules apply to companies outside the EU when the output of their AI systems is used within the bloc.

Deadline to comply

Existing generative AI systems placed on the market before Aug. 2 have until Dec. 2 to comply with the machine-readable marking requirement. The limited grace period does not apply to the other transparency obligations, while content produced before Aug. 2 does not have to be labelled retroactively.

National market surveillance authorities will have primary responsibility for enforcement, alongside the EU’s AI Office in areas under its supervision. Violations can result in fines of up to €15 million ($17.3 million) or 3% of a company’s worldwide annual turnover.

The commission has also approved a voluntary code of practice offering companies a standard route to demonstrate compliance. About 190 companies and organizations had signed the code by the end of July.

Several major technology platforms introduced AI labels before the requirements took effect. TikTok says more than 3 billion videos have been tagged through creator disclosures, detection systems and technologies such as Content Credentials and invisible watermarking.

Meta applies “AI Info” labels to content on Facebook and Instagram when it detects industry-standard signals or receives disclosures from users.

Google has signed the EU transparency code and is developing marking and detection systems using technologies including SynthID and Content Credentials. The company nevertheless warned that overlapping labels and legal notices could create confusion instead of giving users clearer information.