EU AI Act Phase Two: What the July 2026 Rules Mean for Global Tech

Brussels — The European Union's landmark AI Act moves into its most consequential phase this July, with obligations that reach far beyond Europe's borders. Providers of general-purpose AI models must now publish detailed summaries of training data, submit evaluation reports for systemic-risk models, and demonstrate compliance with EU copyright rules.

Regulators say the rules aim to make advanced AI auditable without freezing innovation. Companies including OpenAI, Anthropic, Google DeepMind, Meta, and Mistral have all filed initial transparency reports in the last two weeks. Chinese labs including DeepSeek and Alibaba's Qwen team have opened European representative offices to remain accessible to EU users.

For businesses using AI, the new phase brings clarity but also cost. Enterprises deploying high-risk systems in hiring, credit scoring, education, and critical infrastructure must maintain logs, human oversight, and post-market monitoring. Startups under a certain revenue threshold receive a lighter compliance path via the AI Office's SME sandbox.

Critics argue the disclosure regime still leaves too much wiggle room on data provenance. Rights holders, particularly publishers and music labels, are already preparing test cases against models trained on copyrighted material without explicit licensing. Supporters counter that the Act is the world's first serious attempt to translate AI policy into enforceable law, and that the U.S., U.K., Japan, and Brazil are all watching closely.

Economists at the Bruegel think tank estimate that the compliance market alone will exceed 4 billion euros by 2028, spawning a new tier of AI auditors, red-teamers, and technical governance specialists. For global users, the practical result may be quieter but tangible: clearer labels on AI-generated content, standardized incident reporting when models misbehave, and a European baseline that many multinationals will simply adopt worldwide.