EU creates team to control Artificial Intelligence companies

The European Union (EU) has set up a team, together with national authorities, to control AI companies under the European Artificial Intelligence Regulation (‘AI Act’), which enters into force on Sunday.

31/07/2026
by

Lusa

A from 02 August 2026, the European Department for AI (‘AI Office’), together with national authorities, begins to apply the Artificial Intelligence (IA) Act, on the same date as new transparency rules are being applied, which require certain AI systems to inform users when they are interacting with AI and when the content was generated or changed by it.

With the creation of the AI Office, the European Commission wants to monitor the use of AI models in relation to violations of its new regulation, such as the publication of sexually explicit material, fake photos and videos, and cyber threats to public infrastructure, and the department will be able, as of Sunday, to enforce the rules for AI model suppliers for general purposes (GPIA).

These GPIA models can perform many different tasks and be used in a wide range of tools and services, including AI agents.

It wasn't just OpenAI. Anthropic reveals that IAs attacked other companies

After the case between OpenAI and Hugging Face, Anthropic did an analysis of its own systems and found evidence of three occasions when its models of Artificial Intelligence invaded other organizations autonomously.

Miguel Patinha Dias

The rules also cover more advanced models that may pose systemic risks, and their suppliers must fulfil additional obligations to address risks of large-scale damage, such as those associated with chemical, biological, radiological and nuclear incidents, loss of control, cyber attacks, harmful manipulation and threats to fundamental rights.

According to a communiqué from the Community Executive, the team will also monitor risks that have been attracting public attention recently, such as threats to European cybersecurity and AI’s action outside human control.

Also from Sunday, AI companies are forced to make it clear to consumers, through labels or digital watermarks, that ‘chatbots’ or images were generated by AI.

OpenAI models attacked and infiltrated various platforms

Two OpenAI artificial intelligence (IA) models that escaped autonomously from the controlled environment in which they were tested to attack the Hugging Face site also infiltrated other platforms, the company revealed.

Lusa

Under the new rules, ‘deepfakes’ (images, videos or audio edited or generated by AI) will have to be identified and ‘client support chatbots, virtual wizards or any other conversational interface based on this technology will have to inform users that they are a product generated by AI and not real people.

The entry into force of the rules on high risk AI systems was postponed to 2 December 2027 and the application of the rules on high risk AI systems integrated into regulated products by 2 August 2028.