The Founder’s Guide to AI Policy & Regulation: One year on

When we published the original Founder’s Guide to AI Policy and Regulation in November 2024, the AI policy landscape looked a little different:

• The global conversation around AI policy was still centred on frontier AI model safety, with innovation secondary
• In the UK, the AI Security Institute (then still known as the AI Safety Institute) had just turned one
• The EU AI Act had just been adopted but it was too early in the implementation timeline to assess its impacts
• President Trump had just been elected, promising to repeal Biden-era reporting, safety and governance requirements for AI

Since then, there has been a major narrative shift globally to prioritise AI innovation, driven by the Trump administration in the United States. Governments around the world are now introducing major programmes and reforms to make it easier to develop and deploy AI, driven by increasing concerns about slow economic growth and over-dependence on a handful of US companies. With this comes huge opportunities for founders to capitalise on government support to start and scale their business: ranging from R&D funding and debt and equity financing, to regulatory sandboxes and public compute access.

What has been less loudly championed by governments is growing regulatory divergence: rapid deregulation by the US federal government; a patchwork of AI laws passed by US state governments; uneven EU AI Act enforcement; and still no UK AI legislation on the horizon. For founders, there are now genuine strategic choices about where to build, where to incorporate, and where to focus compliance resources. 

We’ve combed through all the latest developments in AI policy and regulation across each of the guide’s key markets, and drawn out the highlights to bring you up to date and help you cut through the noise.

Essential top tips for founders

All the Top Tips in the Founders’ Guide still stand: good data practices, responsible AI frameworks, compliance and risk assessments are as important now as ever. But here are a few additional ones to keep up with the evolving AI landscape:

1. Capitalise on government support: governments across the UK, Europe, and US are rolling out support to drive development, deployment, and diffusion of AI (think R&D funding, public compute access, access to debt and equity financing, export support). Stay abreast of these, as they provide funding and partnerships at scale.
2. Be clear on the provenance of your training data: copyright reform is one of the largest uncertainties in training AI models. Consider licensing strategies, public domain data, or synthetic alternatives.
3. Invest in compliance: AI regulation is now in force in the EU and certain US states, and as these new laws bed in, enforcement will soon follow. 

Key developments by region

The UK

The UK is trying to position itself as the best place in the world to start and scale an AI business: strongly prioritising innovation and investment with £2 billion announced to support the AI Opportunities Action Plan, and a Compute Roadmap setting out commitments to 20x scaling of UK public compute capacity. It has already launched the AI Research Resource Rapid Access route in July 2025, granting micro, small, and medium sized businesses access to public AI-optimised high-performance computing. More broadly, the government has increased the British Business Bank’s financial capacity by two-thirds to £25.6 billion, with a new mandate to support Industrial Strategy sectors (including AI), and small businesses across the UK.

• Advice for founders: Consider how to take advantage of this support for starting and scaling AI businesses – particularly if your business is R&D-intensive.

To match this, the government is pursuing a light-touch approach to regulating AI, with no AI Bill (yet). The Data (Use and Access) Act passed in July clarified data laws, and once relevant provisions enter into force early next year, will liberalise use of AI in automated decision making. However, the government has yet to resolve a major issue for AI developers in the UK: the copyright regime, and whether the government will pursue an opt-in or opt-out system for text and data mining. We expect this to be included in an AI Bill alongside potential frontier model regulation, possibly coming in 2026 or later.

Finally, the government is seeking to reduce regulatory bottlenecks to applied AI innovation through AI Growth Labs. These Labs will have the power to break down regulator siloes to disapply regulation impeding applied AI innovation, and potentially make permanent regulatory changes. 

• Advice for founders: Copyright represents your biggest uncertainty in the UK, as model developers currently face unclear boundaries on training data. Copyright and regulatory reform could fundamentally shape the UK’s attractiveness for AI development longer term: track this closely and maintain flexibility to adapt your data strategies.

European Union

In Europe, the EU AI Act has sprung to life, as key provisions were implemented in the Spring and Summer: AI systems presenting “unacceptable risk” became prohibited, transparency obligations were placed on “general-purpose” AI, and the penalty regime was activated with fines up to €35 million or 7% of global annual turnover.

As we remarked in April, the EU AI Act now looks increasingly burdensome against this new global drive to innovate – and this time the European Commission seems to agree. After lobbying from Member States and tech companies, the Commission officially announced as part of its “Digital Omnibus” in November that some of the Act’s provisions concerning “high risk” AI models are to be delayed from 2026 to the end of 2027. At the same time, the EU is investing heavily in innovation and business growth initiatives for AI founders and start-ups, including Digital Innovation Hubs, regulatory sandboxes and the recently announced Apply AI Strategy, which allocates €1 billion in EU funds to boost AI adoption and innovation across 10 key sectors. 

• Advice for founders: The 2 August 2026 deadline for high-risk AI systems is a key compliance deadline to watch (if the Commission does not choose to delay compliance timelines) – especially if you’re building AI for employment decisions, credit scoring, critical infrastructure, law enforcement, or education. Also, remember the AI Act’s extraterritoriality; if you aren’t based in the EU but have EU-based users, you’ll need to be compliant.

The United States

The federal US government is squarely prioritising AI innovation and deregulation over safety and risk management.  This is no better exemplified than by the US AI Safety Institute being reformed into the Center for AI Standards and Innovation (CAISI) in June.

President Trump’s AI Action Plan, published on 23 July 2025, is a defining document for federal AI policy, with 103 policy recommendations across innovation, infrastructure, and diplomacy to help the US achieve “global dominance” in AI. Most prominently, it seeks to promote use of open-source and open-weight models, establish regulatory sandboxes to support applied AI innovation, and seeks to remove “ideological bias” from AI risk principles. However, there are major questions about deliverability, at a time when the federal government is also cutting back federal grants and restricting inward talent migration via high H1-B visa fees. 

Copyright is still a live issue in the US: with dozens of lawsuits in federal courts and no clear moves to create federal policy certainty. While Thomson Reuters won against the fair use defence on the basis the training data was not “transformed”, Anthropic partially won its defence against a group of authors on the basis that its use of lawfully acquired books for AI training was “quintessentially transformative”, and thus protected by fair use.

• Advice for founders: Founders training AI models in the US should document training data sources meticulously, and strongly consider licensing strategies, public domain data, or synthetic alternatives.

The locus of regulatory activity continues to be in the States. California, Texas, Colorado, and Utah have passed cross-sectoral AI governance legislation, and 38 states have passed some form of regulation relevant to AI. The federal government is trying to stymie this divergence: the US Congress has tried multiple times to pass a moratorium on state-level AI regulation, and the White House Office for Management and Budget has been empowered by the AI Action Plan to restrict AI funding for states perceived to have onerous AI regulation. 

• Advice for founders: Federal deregulation doesn’t mean regulatory freedom in the US. There are now dozens of state requirements in place, without harmonisation. There are 2 potential options if you operate across state lines:
  • The California Effect – identify the most stringent regulatory standards applicable to your AI product, and apply this to your product nationally
  • Bespoke compliance – consider flexible compliance frameworks that accommodate variation. Consider whether limiting operations to specific states makes strategic sense, or whether your compliance burden justifies the expanded market access.

For more insights, read the full Founder’s Guide to AI Policy & Regulation