AI regulatory horizon scanner
AI regulation is arriving fast and unevenly across the world — the EU AI Act phases in over years, the US mixes a federal risk framework with a patchwork of state laws, and Asia-Pacific economies are each moving at their own pace. This offline reference pulls the major frameworks together by jurisdiction, with status, effective or phased dates, and headline obligations, so you can see the horizon at a glance instead of stitching it together from press releases.
How it works
The data is built into the page — no network calls, nothing tracked. You filter by region (or view everything) and each framework shows its governing instrument, the responsible authority, its current status, the key dates (including phased roll-outs), and a short list of the obligations that matter most. Entries are ordered so the nearest and most binding obligations are easy to spot.
Notes and how to use it
Treat this as an orientation map, not a compliance database. Regulation changes month to month, so confirm specific dates and duties against the official text — the entry names the instrument and authority precisely so you can find the primary source. For planning, filter to the markets you actually operate in, find the nearest effective dates, and map your highest-risk AI systems against the obligations that land first. That turns a sprawling global picture into a prioritised roadmap for your own products.
The major frameworks in brief
EU AI Act — the world’s first comprehensive AI regulation, built on a risk-tiered system. Prohibited practices (banned outright) and general-purpose AI obligations activated first, with the bulk of high-risk requirements phasing in over a two-year window. High-risk AI systems — those used in employment, credit scoring, critical infrastructure, law enforcement, education, and other sensitive areas — face conformity assessments, technical documentation, and registration in an EU database. General-purpose AI models (the large foundation models) have their own transparency and copyright obligations.
UK approach — the UK has chosen a principles-based, sector-led approach rather than a horizontal statute. Each regulator (the FCA, ICO, CQC, Ofsted, etc.) applies the government’s five cross-sector principles — safety, security, fairness, accountability, contestability — within its own domain. The result is less uniform than the EU AI Act but more flexible. The government has committed to monitor and potentially legislate if voluntary approaches prove insufficient.
US federal landscape — the US operates through a patchwork of agency guidance, executive orders, and the voluntary NIST AI Risk Management Framework. There is no federal AI statute equivalent to the EU AI Act, though sector agencies (the FDA, CFPB, EEOC, FTC) are applying existing authority to AI. Several states have passed or are advancing their own AI legislation, with California and Colorado most active.
China — China has taken a targeted regulatory approach, with specific regulations on generative AI, algorithm recommendation systems, and deep synthesis (deepfakes) rather than a single omnibus law. Generative AI services available to the public in China must register with the Cyberspace Administration of China and comply with content standards.
Singapore and other Asia-Pacific — Singapore’s Model AI Governance Framework is voluntary but widely referenced as a practical template. Japan, South Korea, and Australia are each developing their own approaches, generally lighter-touch than the EU but moving toward clearer expectations for high-risk applications.
Planning under regulatory uncertainty
The most common mistake organisations make is waiting for final text before acting. By the time a regulation is fully in force, the window for compliant design has usually closed — retrofitting transparency, human oversight, and documentation into a shipped system is far more expensive than building them in. Use the horizon scanner to identify the nearest obligations, check whether your highest-risk systems fall into any high-risk category under the applicable regime, and begin the documentation and governance work now. The cost of being early is low; the cost of being late is not.