EU introduces AI transparency requirements to curb manipulation

New artificial intelligence transparency requirements entered into force in the European Union in August, ERR reports, requiring people to be informed when they are interacting with an AI system or viewing content created or altered by AI.

The rules aim to harmonise obligations across sectors and protect people from manipulation involving AI. They are intended to help users identify when they encounter AI-generated content or communicate directly with an AI system.

Aivar Pau, strategic communications adviser for digital affairs at Estonia’s Ministry of Justice and Digital Affairs, said the regulation was the EU’s first cross-sector AI regulation. Earlier rules had applied to specific sectors, while the new AI regulation harmonises requirements across fields, he said.

Under the new requirements, AI systems must indicate that users are not communicating with a person. Specific labels are also required for deepfakes and other content created or modified with AI that could be mistaken for authentic material.

Exceptions to notification requirements

A separate label is not required when the use of AI is self-evident. Pau said this could be the case when a user enters an environment clearly presented as an AI chatbot or an artificial intelligence character, or when the service’s name and design clearly indicate that it uses AI.

However, the use of AI may not be obvious when it appears to act on behalf of a person or in a human-like manner without a clear indication. Users may then mistakenly believe they are communicating with a real person, or fail to notice that AI has been integrated into a service.

People must also be informed when they encounter AI systems capable of processing biometric data and identifying or inferring emotions or intentions, or placing people into categories based on characteristics such as gender, age, appearance, ethnic origin or personal preferences.

The Ministry of Justice and Digital Affairs said this information must also be provided in a format accessible to people with disabilities.

Notification is not required when the use of AI is legally permitted for detecting, preventing, investigating or prosecuting crimes. In such cases, safeguards must be applied to protect the rights and freedoms of third parties, Pau said, except when people use the systems to report crimes.

Exceptions also apply to edited texts. AI assistance used in preparing published text does not have to be separately disclosed if the author, publisher or editorial staff has reviewed the text.

In that case, the individual or legal entity responsible for publication bears editorial responsibility for the AI-generated content, the ministry said.

AI labels and supervision

The European Commission has developed three visual labels for indicating AI use: a basic icon, a label for content created entirely by AI and a label for partial AI use. Under the new law, the labels must remain with the text throughout the creative process.

The EU’s labels are available through the European Commission’s digital policy portal. The new requirements apply both to providers of AI systems and to their users.