Think of all the things you need to do your job that you have little control over. What happens when those things are suddenly unavailable, or the prices shoot up, or reveal a specific bias? This is what AI sovereignty addresses and it’s a growing movement.
You’ve heard of sovereign AI, i.e. open source LLMs and data that is localised in order to provide a perspective that is independent from larger mainstream models. But AI sovereignty is more than that; it’s a tech-cultural evolution that is manifesting itself in many forms. Here, Artificial Lawyer looks at some of the ways this ‘sovereignty’, this drive for independence and the ability to control one’s own use of AI tools, is taking shape.
Your ‘Own’ LLMs
The first and most clearly outlined example is ‘sovereign AI’ – (which to AL is a subset of the wider movement) – and as noted, refers to the idea that if you want to escape dependence on what are largely US-based LLMs, e.g. OpenAI, Anthropic, and Gemini, and in turn the regulatory capture and technological supply chain that governs them, then you have to use LLMs developed, or trained, in your region.
These issues were explored by several speakers at Legal Innovators Europe in Paris last week, including Legora’s France chief, Jonathan Williams, and Noxtua from Germany. They highlighted not just the reality that the US Government can cut the world off from using certain models, but there is also a complex supply chain of data and chips which are also largely dominated by the US.
In short, something that will become increasingly vital to global businesses – and the lawyers that serve them – is far from being a controllable resource. And that understandably worries the lawyers who are ‘out of the loop’.
As AL heard at the conference, for some the answer seems to be efforts to regulate their way out of their lack of control. But, that seems unlikely to succeed, no more than taxing successful people makes a country’s economy grow faster. Another, more sensible path, is to build out your own LLMs (e.g. France’s Mistral), or tap multiple other open source LLMs and post-train them to meet your specific needs, perhaps on your own data sets that represent the areas of the law you care most about.
AI Sovereignty is More Than Geopolitics
However, the more Artificial Lawyer has dwelled on this subject, the more it’s clear that geopolitics and sensitivities over cultural centrality of AI training data is really just one expression of a wider movement that seeks to gain some control over the most powerful knowledge-based technology since the arrival of the printing press.
And, it should be added, that as token costs become more of an issue as well, there is a clear economic aspect here that will only grow.
Here are some examples –
Thomson Reuters training its own open source LLMs – at first glance you may not see this as AI sovereignty, but to AL this is a classic expression of the urge to control, at least partly, one’s own AI destiny. TR has a huge amount of legal data. Why just be dependent upon OpenAI or another company to provide the foundation layer of what you offer your customers? TR can build an independent – sovereign AI capability – grounded in its own curated data. It could, potentially, also save the company plenty of money on token costs. If at some point in the future the major model makers were curtailed even more than is happening today, e.g. Anthropic’s Fable getting temporarily banned, TR can still operate as it wishes to.
Kirkland & Ellis apparently planning to build its own GPU clusters for open source LLM training (see AL story) and working with Palantir to model key workflows. Again, this may not immediately make you think of sovereignty, but it very much is all about that. Kirkland – one of the most profitable law firms on the planet – wants to control the means of production when it comes to legal AI. It made a very vocal point about not using well-known platforms. It wants to build ‘its own thing’. In this case, it’s not about avoiding US models, it’s about controlling not just their own data, but controlling their own AI journey – and, let’s be frank, their own narrative, which includes a major dose of marketing messages too. And there is nothing wrong in that. Plus, this may prove to be a cost-saving move as well.
Harvey working with selected law firms, and potentially some of their clients, on open source model training. The reasoning is similar to above. Law firms want to do their own thing. They want exclusivity. They want to bottle what is special. (Note: what can truly be unique may be the client relationship itself – see here.) But, again, this is an expression of sovereignty. What many lawyers don’t want is to be wholly dependent on something that is A) uncontrollable, and B) has no specific connection to them, and therefore dilutes their value.
Vibe-coding, MikeOSS and LegalQuants are also an expression – to AL at least – of AI sovereignty. The lawyers who take tool production into their own hands, albeit using the main LLMs, are still saying the same thing as those above: ‘We want to be in control. We don’t want to be dependent upon X or Y supplier. We can do our own thing. We can build our own AI path to the future of the legal profession.’ In this case, the sovereignty goes right down to a personal level – a one-man army, if you will.
Conclusion
Every action has an equal and opposite reaction. It’s only natural that as a handful of LLM makers in California become increasingly powerful and vital for our ability to run our lives and our businesses – and in turn they become susceptible to regulation and supply chain risks – that many people want a way out; a way to take back some control.
If every book in the world had to be printed by one printing press and edited by one editor, that would scare people. The same is true for AI. People tend towards freedom. The legal profession – despite what some may think – is all about liberty and independence. A single lawyer can take on the largest corporation or government. That person’s wits, the law itself, and the desire to act, are sufficient to make a stand. When that lawyer starts to feel that AI is potentially going to undermine that independence they instinctively feel the need to react.
And AL would say: they are right to do so. AI can be a liberating force for all of us. A printing press, a global library, a second brain, as it were, in our pocket, as well as a powerful ‘means of production’. It really is an incredible opportunity for everyone who seeks liberty and independence. Equally, it can be regulated, banned, and formed in such a way that we become subject to it, rather than users of a tool.
One could say that our relationship with AI is only going to become more complicated and more vital as time goes by, and AI sovereignty – in its broadest sense – will become an increasingly global movement both within and without the legal world.
Richard Tromans, Founder, Artificial Lawyer.
—
(Conference advert)
Come and join us in New York and London this November at Legal Innovators!
Legal Innovators New York – Nov 17 and 18.
Discover more from Artificial Lawyer
Subscribe to get the latest posts sent to your email.