New obligations under the European Union’s Artificial Intelligence Act came into force on 2 August 2025, marking the first major compliance milestone for providers of general-purpose AI models. From this date, companies that develop or deploy large models must maintain detailed technical documentation, publish summaries of training data, and follow guidance on copyright, safety testing and post-deployment monitoring. The legislation applies to firms worldwide that make AI services available in the 27-nation bloc. Non-compliance can draw fines of up to €35 million or 7 percent of global annual revenue, while lesser infringements carry penalties of up to €15 million or 3 percent. A one-year grace period means enforcement actions cannot begin before August 2026, but companies must already show good-faith progress. Several major vendors—Amazon, Google, IBM, Microsoft, Anthropic, OpenAI and others—have signed a voluntary Code of Practice intended to ease eventual compliance. Google said it still fears the rules could slow European innovation, and Meta has refused to sign, calling the framework “overreach.” Uncertainty over the Act’s scope is disrupting product roadmaps, particularly for start-ups, according to technology lawyer Oliver Howley of Proskauer. He warns that ambiguous definitions of what constitutes a GPAI “provider” and the risk of exposing trade secrets in data summaries are causing some firms to delay EU launches or restrict access.