The EU AI Act entered a critical phase this week, with compliance deadlines approaching for general-purpose AI models (GPAI). Providers must meet obligations by August 2, 2025, though models launched before this date have until 2027 to comply. The European Commission clarified GPAI scope through new guidelines, addressing concerns about extraterritorial enforcement and training data requirements. These rules apply to high-risk applications like healthcare and hiring, requiring robust documentation and human oversight.

In the UK, the Data (Use and Access) Act 2025 became law, modernizing data-sharing frameworks and updating privacy rules. While it doesn’t explicitly regulate AI training data, debates continue about copyright protections for AI model development. The UK government is also advancing a proposed AI Bill, emphasizing sector-specific regulations over broad restrictions.

Across the Atlantic, the US saw a pivotal shift when Congress rejected a 10-year moratorium on state AI regulations. This allows states to enforce their own rules, though federal funding may still influence compliance. The Trump administration countered with an AI Action Plan prioritizing deregulation, infrastructure investment, and 'ideologically neutral' AI systems. Three executive orders target federal procurement practices, restricting AI systems deemed 'woke' and revising NIST’s risk management framework to remove DEI references.

Globally, the UN advanced a draft resolution urging member states to adopt AI governance frameworks. Countries like Taiwan, Turkey, and the UAE are developing sector-specific guidelines, while the UK advocates flexible, industry-tailored approaches. Meanwhile, 93% of financial institutions plan to adopt agentic AI by 2027, driven by regulatory pressures and efficiency gains. These systems face scrutiny under emerging frameworks, particularly in high-risk categories like finance and healthcare.