Takeaways From UC Berkeley Law AI Institute

Last week I had the pleasure of attending UC Berkeley Law AI Institute, where we heard from academics, VCs, judges, private practice lawyers, GCs for the biggest AI companies and other companies like life sciences and the creative industries. The agenda spanned IP, privacy, European regulation and more. I've avoided naming speakers, since many disclaimed that they spoke in their personal capacity, but if you want to tag yourself, please feel free. My top 5 takeaways:

1. AI Governance is not keeping pace with adoption, especially in law firms
   We heard that only 9% of law firms have a written AI policy that is actually enforced, whilst 43% have no formal AI policy and no plans to create one. 54% of law firms have no AI training and no plans in place to launch some. This is truly shocking, given the sensitivity of information and use cases by law firms of AI. What needs to happen for this to change? Will the industry wait for the first lawsuit, cyber incident or regulatory fine?
2.  Courts are a forgotten stakeholder in AI implementation
   We heard from the judiciary and court executives about the need for dedicated AI purpose-built for the courts and there was a question around whether the rise in self-represented litigants is caused by AI. There was some good discussion about whether a judge should use AI and when they should disclose the use to the participants. Would you, for example, be comfortable with a judge in a close evidentiary ruling asking AI for the strongest and weakest arguments for admitting certain evidence? Should a judge be able to use AI to stress test their reasoning? What about to compare sentencing in similar cases or check a witness' testimony against the record, or make a simplified alternative order in plain language? Not straightforward questions and an often missed element of the “AI in law” debate.
3. The Regulatory Landscape is fragmented and still forming
   Whilst the EU plows ahead with its EU AI Act, there are more than 1400 proposed AI state bills in the US. GCs need to help their firms navigate an evolving, and potentially, contradictory landscape. For early stage GCs, is there a possibility your clients will accept the majority of legal risks and mature their response to risk as they get more funding and potentially at the behest of their investors?
4. The Funding of AI in the US is huge, and focussed on the model layer
   86% of US venture capital dollars went to AI in H1 2026 and 11% of companies in AI overall get 67% of the value. In fact, 6 frontier labs get 97% of the funding. There's a big focus on funding the model layer, but in fact the application lab needs less funding because the app developers can rely on the expensive training and efforts of the model layer. There are starting to be fewer deals, but with much bigger cheques. 
5. Law is straining against AI
   Trade secrets were hot on the agenda and their future role in big litigation involving AI. There are still questions about how IP and privacy are applied to AI. I hear a number of people say AI companies were moving away from a zero data retention policy, which is really huge for AI governance. Will this push companies towards novel AI insurance and how far will we see existing contracts or clauses like “your data may be used to improve our services” stretched to serve a new AI business model. Is there a privacy dimension to outsourcing our cognitive load to an AI tool, if privacy is really all about autonomy? GCs exploring AI risk in the context of their business need to understand their specific risk. For example, in the creative industries, an extra finger is a quality issue, in a contract an extra word is worth millions.