EU Intensifies AI Oversight, Targets OpenAI, Anthropic Under New AI Act Rules

The stakes have been raised for U.S. businesses like Anthropic, OpenAI, and Google as the European Union has acquired additional powers to scrutinise AI models. The EU now limits access to EU markets and penalises model providers. With the new powers that were implemented on August 2nd, the EU executive arm, the Commission, can now request model evaluations prior to their public distribution in the region.

This means that providers will have less access to the EU market and could face fines of up to 15% of their yearly revenue or 15 million euros, whichever is greater. With the expanded powers come further potential disagreements between the United States and the European Commission, which has already been at odds over European demands for technological autonomy and penalties imposed on American technology corporations.

Tussle Between EU and US Grows

In July, European regulators accused Google of favouring its own services, leading to a $1 billion fine. As a result, President Trump of the United States threatened the European Union with a "substantial" tariff. Part of the phased implementation of the EU AI Act in 2024 is the transfer of supervisory and enforcement authority over general-purpose AI models to the EU AI Office. Some parts of the law will be in transition for a while because of the Act.

A statement was made by Henna Virkkunen, the European Commission's executive vice-president for tech sovereignty, security, and democracy, warning that AI, even the most sophisticated models, pose significant threats if not used correctly. New animosity has emerged between AI research facilities and the European Union as a result of the capability of AI recently. Prior to Anthropic's announcement that it would share access, the bloc had been requesting access to the Mythos model for months. Europe has recently taken steps to strengthen its technological independence in response to growing tensions with the Trump administration and the need toposesuce reliance on American technologies.

EU’s Code of Practice

In addition to implementing the rules, the Commission also published the initial roster of more than 180 groups that had agreed to the AI-generated content transparency Code of Practice. To provide businesses with a tangible means of demonstrating compliance with the labelling and marking regulations, it offers a voluntary framework.

While participation in the Code of Practice is entirely optional, the transparency duties it enshrines are mandated by law (Article 50) regardless. By signing, providers can provide evidence of compliance in a written form.

The pace of progress on certain sections of the AI Act varies. The AI Omnibus, a collection of amendments to the Act, has postponed the regulations for high-risk AI systems until December 2, 2027, and those for high-risk systems integrated into regulated products until August 2, 2028. Automated systems that create sexually explicit content without consent or content that depicts sexual abuse of children will be illegal from December 2, 2026.