Skip to content

Bringing you global stories from a neutral view

Technology

EU AI Act Provisions Take Effect, Giving Brussels New Powers

Key provisions of the EU's AI Act took effect Sunday, banning AI manipulation, mandating deepfake labels, and giving Brussels authority to audit major AI models.

EU AI Act Provisions Take Effect, Giving Brussels New Powers

Several provisions of the EU’s AI Act took effect Sunday, granting Brussels new powers to scrutinize artificial intelligence models and penalize companies that break the rules.

The law, adopted two years ago, bans using AI to deceive or manipulate people or exploit vulnerabilities linked to age or disability. It restricts biometric data and facial recognition in AI systems, and requires companies to assess and mitigate the risks their models pose.

Transparency requirements

New transparency obligations also started Sunday. AI systems must watermark the content they generate so it can be easily identified, and new deepfakes must carry labels such as icons or badges. Models already on the market have until December 2 to comply.

The AI Office, a body within the European Commission, can now evaluate large AI models and require providers to grant access, in some cases before a product reaches the market. National regulators are responsible for smaller models. Brussels can also investigate companies, demand corrective measures, and restrict the deployment of new models where violations are found.

An anonymous Commission official said Friday that the EU now has the power to conduct in-depth evaluations and would use it as much as necessary, declining to comment on any specific company. The source notes the EU has struggled in recent months to access advanced models developed by foreign AI companies, citing Mythos from American group Anthropic as an example.

Fines and future rules

Companies found to have carried out banned activities face fines of up to 7 percent of global annual turnover or 35 million euros, whichever is higher. Other violations carry penalties of up to 3 percent of turnover or 15 million euros, with additional fines for obstructing Commission investigations.

Further rules are on the way. AI systems capable of generating non-consensual nude images will be formally banned from December. Separate rules covering high-risk AI in sensitive fields such as health and safety will apply from 2027 and 2028, after their implementation was recently delayed to give companies more time to prepare.

Related

Leave a comment

Your email address will not be published. Required fields are marked *