Sponsored by
This is a quieter, more reflective week, and the thread running through it is control: who owns the models, the data and the know-how underneath the AI you use. The lead is the growing "AI sovereignty" conversation, the idea that depending on a handful of mostly American model providers is itself a risk worth managing. It is an abstract-sounding topic with some very practical questions attached, and this issue tries to draw those out.
AI in Practice
AI sovereignty, and what it means for a firm that will never build its own model
Artificial Lawyer published a piece on 29 June on what it calls "AI sovereignty", picking up a theme that ran through Legal Innovators Europe in Paris the week before. The short version: as a small number of mostly US-based model providers (OpenAI, Anthropic, Google) become central to how businesses and their lawyers work, some organisations want a way to reduce their dependence on them. The drivers are geopolitical and practical at once: the worry that access to models, or to the chips behind them, could be restricted, the steady rise in token costs, and a growing unease about pouring your hard-won know-how into a platform you do not control.
This is particularly true in light of recent US government intervention restricting access to GPT 5.6 and Fable 5. What happens when you build workflows on these models and they are restricted, retired or otherwise inaccessible to you?
The responses sit on a spectrum. At the top of the market, the moves are very ambitious: Thomson Reuters says it is training its own models on its own curated legal content, and Kirkland & Ellis is assembling its own stack, with its own compute, a Palantir deal, and, this week, an exclusive arrangement to build proprietary tools on the litigation platform Syllo. Harvey is reported to be working with firms to post-train open models on their workflows. At the other end, individual lawyers are building small tools for themselves. Different scales, same instinct: we want to be in control.
For the overwhelming majority of UK firms, none of this means building a model, and it is worth being clear about why. Running open or self-hosted models in-house is not yet a serious substitute for the frontier cloud models on real legal work, so "sovereignty" for an ordinary practice is a more modest and more useful idea. It means knowing where your data is processed and stored, understanding what happens if a model is withdrawn or repriced (a real risk, given that Microsoft briefly blocked Claude Fable 5 for its own staff over retention terms, covered here on 12 June), and noticing whether you are quietly handing your best precedents and playbooks to a supplier you could not easily leave.
This author's take is that sovereignty is better read as a checklist. You do not need to build your own model to ask the dependency questions, and you do keep more control than you might think by holding your own data, precedents and know-how in a form you could take elsewhere. The firms that come off worst here will not be the ones that failed to build their own tools, but the ones that never asked what they would do if the supplier changed the terms.
Read: Artificial Lawyer
On your radar
Thomson Reuters opens early access to a rebuilt, "fiduciary-grade" CoCounsel: Thomson Reuters has begun rolling out early access, in the United States first, to a ground-up rebuild of its CoCounsel Legal assistant, which it pitches against what it calls a "fiduciary-grade" standard (grounded in authoritative content, with verifiable outputs) and which, according to LawSites, is built on Anthropic's Claude Agent SDK. General availability in the US is expected in August, with the UK among the markets due to follow. Why it matters for UK lawyers: many firms already license Westlaw and Practical Law, so this will arrive inside tools you have rather than as a fresh purchase, and "grounded and verifiable" is a fair benchmark to hold any vendor to. If you use CoCounsel, ask Thomson Reuters when the new version reaches the UK and test it on non-client work before anyone relies on it. (Thomson Reuters, LawSites)
Kirkland keeps building its own litigation "secret sauce": Kirkland & Ellis has signed a multiyear partnership with the litigation platform Syllo that gives the firm exclusive rights to build its own proprietary tools on top of Syllo's case management, eDiscovery and drafting system. It follows the firm's Palantir deal and its move to build its own model infrastructure, all pointing the same way: the most profitable firms increasingly want to own their AI rather than rent it. Why it matters for UK lawyers: almost no firm can match this spending, but the strategic question it raises applies to everyone: where does your firm sit on build versus buy, and what is the plan for the know-how that makes you distinctive? Put that question on the agenda for your next partnership or innovation meeting rather than letting it drift. (Artificial Lawyer)
Reed Smith puts its partners through an AI course: Reed Smith has launched an "AI Leadership Program" at Cornell University, running its partners through foundational AI knowledge, applied strategy, ethics and governance, with the stated aim of helping them advise clients on AI adoption. Artificial Lawyer raises a reasonable question: whether client value really sits in partners advising on AI deployment, rather than in the ethics and governance points that are squarely legal work. Why it matters for UK lawyers: partner-level AI literacy is becoming a differentiator, and clients increasingly expect their advisers to understand the tools shaping their businesses. You do not need an Ivy League course to make a start: a structured internal session on AI ethics, governance and the firm's own supervision rules would move most teams further than another tool trial. (Artificial Lawyer)
Ad Break
In order to help cover the running costs of this newsletter, please check out the advert below. In line with my promises from the start, adverts will always be declared.
What’s next is almost here.
On July 16th at 1PM ET, beehiiv is going live with a look at the future of publishing, audience growth, and digital business.
What started as a newsletter platform has evolved into something much bigger: a place where creators and brands can grow, monetize, and own their audiences without stitching together half the internet to make it work.
The next chapter starts live at the Summer Release Event.
Join us to see what’s coming next.
For Review
The Platform vs Specialist Debate is Asking the Wrong Question (Artificial Lawyer)
A sponsored piece by Katja Nikolaus of JUNE, and worth reading past the vendor framing for a useful argument: the platform-versus-specialist choice misses the point for high-volume, repeatable legal work (mass claims, data-incident requests, regulatory waves), where the hard part is visibility across thousands of near-identical matters rather than depth on any single one. A helpful lens for in-house teams and any firm with a volume practice.
Read or listen: Artificial Lawyer
AI Hallucination Cases Database (Damien Charlotin)
Staying in control of AI ultimately comes down to verification, and this free, searchable database tracks cases worldwide where AI-generated material (invented citations and the like) reached a court, now including more than 60 from the UK. It is a sobering training resource: circulate a few of the UK entries to fee earners as concrete examples of how quickly unverified output becomes a professional problem.
Read or listen: Damien Charlotin
Practice Prompt
Try the below prompt to interrogate an AI tool your firm depends on for the "sovereignty" questions this week's lead raises: where your data lives, what happens if the model changes, and whether you could ever leave. Ensure you fill in context and constraints and other aspects marked with {}. Remember to adhere to the Golden Rules and do not upload confidential or privileged information to public tools.
You are a legal technology procurement assistant. Your task is to help a UK law firm
assess how dependent it is on a particular AI tool, and how much control it retains, as
a planning aid only.
The tool:
- Tool and provider: {e.g., "Harvey" / "CoCounsel (Thomson Reuters)" / "Claude for Legal"}
- What the firm uses it for: {e.g., "first-pass contract review" / "legal research" /
"drafting" / "case summarisation"}
- How it is accessed: {e.g., "cloud SaaS" / "API" / "embedded in Westlaw or Practical Law"}
- What the firm feeds into it: {e.g., "precedents and playbooks" / "client documents" /
"no client data"}
Produce a "dependency and control" review under these headings:
1. Underlying model and change risk
- Which model or models sit under this tool, and can the provider change them without notice?
- What would a model change, withdrawal, or performance drop mean for the firm's workflows?
- Is there any commitment on notice of model changes, or the ability to stay on a version?
2. Data location and retention
- Where is data processed and stored (jurisdiction)?
- What are the retention terms, and can prompts or outputs be used for training?
- Which foreign laws could compel disclosure of the firm's data?
3. Cost exposure
- How is the firm charged (per seat, per use, token-based), and what governs price changes?
- What is the exposure if underlying model prices rise, and is there a cap or notice period?
4. Know-how and lock-in
- Is the firm feeding its own precedents, playbooks or know-how into the tool?
- If so, can that be exported, and does the provider gain any rights over it?
- How hard would it be to move to another tool: what would the firm lose?
5. Exit and continuity
- What happens to the firm's data if the provider is acquired, fails, or changes terms?
- Are there adequate termination, data-return and deletion provisions?
- What is the firm's fallback if this tool became unavailable tomorrow?
6. Verdict
Rate the firm's dependency on this tool as LOW, MEDIUM or HIGH, list the two or three
changes that would most reduce lock-in, and identify the questions to put to the vendor
in writing.
Constraints:
- {Add firm-specific constraints, e.g., "data must remain in the UK or EEA" / "we cannot
accept training on our data" / "we must be able to exit within 30 days"}
- Apply the law and regulatory framework of England and Wales throughout.
- Do not invent product features, contractual terms, or provider policies. Where the
position is not publicly known, flag it as a gap and say what to ask the vendor.
- This is a procurement planning aid, not legal advice.How did we do?
Hit reply and tell me what you would like covered in future issues or any feedback. I read every email!
Thanks for reading,
Serhan, UK Legal AI Brief
Disclaimer
Guidance and news only. Not legal advice. Always use AI tools safely.
Recommended Newsletters
Below are a few newsletters that I recommend, for various reasons. Check them out!
