In partnership with
After a month dominated by judges admonishing the profession for AI failures, the government has made announcements. At London Tech Week on 9 June, David Lammy announced that legal services will be the first sector admitted to the government's new AI Growth Lab, a regulatory sandbox bringing the SRA, the ICO, the Legal Services Board and the Council for Licensed Conveyancers around the same table. Elsewhere, Anthropic released Claude Fable 5, the most capable legal AI model yet measured, and within two days Microsoft had blocked its own staff from using it over data retention concerns. Both stories are below.
Legal services named first sector in the government's AI regulatory sandbox
Speaking at London Tech Week on 9 June, the Deputy Prime Minister and Lord Chancellor David Lammy announced that legal services will be the first sector to join the government's AI Growth Labs, an advisory regulatory sandbox designed to help organisations develop and deploy AI products within existing regulatory frameworks. The Ministry of Justice said applications will open later this summer for legal services providers, legal tech firms and conveyancing companies, with other sectors to follow later in the year.
The Lab brings together the Solicitors Regulation Authority, the Legal Services Board, the Information Commissioner's Office and the Council for Licensed Conveyancers. The stated purpose is to give participants a safe space to test new legal services products and discuss regulatory issues directly with the regulators, with clearer, joined-up guidance within existing rules. It is worth being precise about what this is not: the announcement describes an advisory sandbox, not a statutory regime. No rules are being switched off, and participation will not suspend any SRA or data protection obligation.
The same speech contained a second announcement: AI legal assistants for Crown Court professionals (supporting routine casework, research and case analysis) and a separate AI tool to help judges identify trial-ready cases and group similar hearings (Computer Weekly). The context is a Crown Court backlog that stood at 80,200 cases at the end of December 2025, the highest on record, with figures obtained under freedom of information laws showing trials listed for 2028 and later. The Ministry of Justice says the technology will be trialled against standards set by judges and lawyers before any rollout, and the Law Society has already called for the results of all AI trials to be published and assessed before wider implementation, which seems the minimum that should be expected. Criminal practitioners should follow those trial standards as they are published, and firms with criminal practices would do well to start discussing now how AI-produced case summaries will be verified before anyone relies on them.
The shift in tone matters as much as the substance. The past two months of this newsletter have been dominated by judicial criticism and SRA referrals. This announcement is the other half of the regulatory picture: the same regulators who are dealing with AI misuse are now offering a structured route for firms that want to do it properly. This author would caution against reading the Lab as a softening. The enforcement record from May has not gone anywhere, and a firm that experiments inside the sandbox while its fee earners use unapproved tools outside it has gained nothing.
Read: GOV.UK | SRA | Law Gazette | Legal Futures
On your radar
Microsoft blocks Claude Fable 5 for its own employees over data retention: within two days of Anthropic launching Fable 5, Microsoft's information security team completed an emergency review and blocked the model across all internal Copilot instances. The reason: Anthropic's data retention policy for Fable 5 retains prompts and outputs for 30 days on every platform, and for up to two years if flagged by its trust and safety classifiers. All other Claude models remain available to Microsoft employees because they operate under zero data retention terms. The irony is difficult to miss: Microsoft rolled out Fable 5 to its paying GitHub Copilot and Azure Foundry customers on the same day its own infosec team concluded the data retention terms were unacceptable for internal use. Why it matters for UK lawyers: if Microsoft's own security team considers 30-day data retention a risk for its confidential information, law firms should ask themselves the same question. Check whether your firm's AI contracts specify data retention periods, and whether those periods are compatible with your confidentiality obligations. This is particularly relevant for firms using Claude through third-party platforms, where the retention terms may differ from direct API access. (PYMNTS)
Claude Fable 5 released: more capable, twice the price: Anthropic released Claude Fable 5 on 9 June, the first publicly available model from its new Mythos class, positioned above Opus in capability. Harvey reports that Fable 5 now tops its open-source Legal Agent Benchmark with an all-pass score of 13.3% across more than 1,200 realistic legal tasks, up from Opus 4.8's 10.4%, with GPT-5.5 at 2.1% and Gemini 3.1 Pro at 0.0%. Two readings of that number are available, and both are correct: this is the most capable legal AI model yet measured, and even it fails the strict end-to-end standard on roughly seven of every eight tasks. Pricing is $10 per million input tokens and $50 per million output tokens, double Opus 4.8's rates, which reinforces the token-cost trend covered in last week's issue. Why it matters for UK lawyers: the model underneath Claude for Legal, Harvey and other tools changes without your procurement team being asked, and this one arrives with a doubled price tag. Ask your vendor which model your firm's tools now run on, what the budget impact of an upgrade would be, and re-test your critical workflows (citation verification above all) before relying on them in anything court-facing. (Anthropic | Harvey | Artificial Lawyer)
Edinburgh's Wordsmith raises $70 million to move legal work in-house: the Edinburgh-headquartered legal AI company closed a $70 million Series B on 3 June, led by Highland Europe and Index Ventures, taking total funding to around $100 million. Wordsmith builds an AI legal operations platform for in-house teams, is used by more than 500 companies (including BT, the Financial Times and Canva). Its pitch is explicit: bringing legal work back in-house and away from law firms. The round came in a busy funding week, with Billables AI raising $10.2 million and Sandstone raising $30 million for its AI-native in-house legal platform. Why it matters for UK lawyers: this is a UK-headquartered company being funded specifically to reduce what in-house teams send to external counsel. Review which of your routine instructions (first-pass contract review, policy queries, standard documents) sit in that exposed category, and consider where your advice adds value that an in-house platform cannot replicate. (Artificial Lawyer | Scottish Legal News)
Kirkland & Ellis partners with Palantir for a PE fund formation platform: following last week's coverage of Kirkland's $500 million AI investment, the firm has launched a specific product in partnership with Palantir Technologies: a "fund formation engine" which builds Kirkland's proprietary knowledge of fund formation documents into an automated workspace. Over 1,000 Kirkland lawyers will use the platform for private equity fundraising, investor compliance and fund management work. Why it matters for UK lawyers: the Kirkland approach (building bespoke tools on its own data, with a government-adjacent tech partner) is one that very few firms could replicate, but the choice of Palantir signals that data security and sovereignty are priorities at the top of the market. Raise with your managing partner or innovation lead whether the firm's AI strategy accounts for the performance gap these bespoke tools may create. (Artificial Lawyer | Bloomberg Law)
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.
Become An AI Expert In Just 5 Minutes
If you’re a decision maker at your company, you need to be on the bleeding edge of, well, everything. But before you go signing up for seminars, conferences, lunch ‘n learns, and all that jazz, just know there’s a far better (and simpler) way: Subscribing to The Deep View.
This daily newsletter condenses everything you need to know about the latest and greatest AI developments into a 5-minute read. Squeeze it into your morning coffee break and before you know it, you’ll be an expert too.
Subscribe right here. It’s totally free, wildly informative, and trusted by 600,000+ readers at Google, Meta, Microsoft, and beyond.
For Review
"Illustrative Use Cases for the Legal Services AI Growth Lab" (GOV.UK)
A short PDF published alongside the Growth Lab announcement, and the most concrete indication yet of what the sandbox is for. It contains three case studies: Garfield.Law, the SRA-authorised AI-driven firm recovering small debts through the small claims court; a conveyancing firm wanting to test an AI tool that flags issues in sales packs (with the CLC); and a large firm with questions about using historical client files to fine-tune an in-house AI tool, answered jointly by the SRA, CLC and ICO rather than by each regulator separately. That third scenario, coordinated answers on client data and model training, is the one most firms will care about. Worth reading before deciding whether to apply.
"The Best AI Model Ever Built Just Landed in Your Claude Account. Don't Let Your Firm Touch It Yet." (Smith Stephen)
A useful counterweight to launch-week enthusiasm. The piece argues that however capable Fable 5 is on benchmarks, firms should not switch production workflows to a model released this week, and makes the case for staged adoption: test on non-client work first, compare outputs against your current model, and only then promote it into client-facing use. Read alongside the Microsoft story above, the argument has aged well within its first 48 hours.
Read or listen: Smith Stephen
Practice Prompt
Try the below prompt to assess whether an AI tool your firm is considering meets your regulatory and data protection obligations, which feels apt in a week when Microsoft's own security team blocked a model over its retention terms. 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 compliance assistant. Your task is to help a UK law firm assess whether a specific AI tool meets its regulatory and data protection obligations before adoption.
AI tool being assessed:
- Tool name and provider: {e.g., "Claude for Legal (Anthropic)" / "Harvey" / "Microsoft Legal Agent" / "CoCounsel (Thomson Reuters)"}
- Deployment model: {e.g., "cloud-hosted SaaS" / "on-premise" / "API integration" / "browser-based" / "embedded in existing software"}
- Intended use at this firm: {e.g., "contract review and redlining" / "legal research" / "document drafting" / "due diligence" / "case summarisation"}
- Who will use it: {e.g., "all fee earners" / "litigation team only" / "paralegals for first-pass review" / "partners for strategic analysis"}
Firm details:
- Firm size and type: {e.g., "6-partner high street practice" / "50-lawyer regional firm" / "200-lawyer City firm"}
- Regulator: {e.g., "SRA" / "CLC" / "BSB"}
- Current AI policy: {e.g., "formal written policy" / "informal guidance" / "no policy"}
- Data sensitivity: {e.g., "we handle legally privileged material daily" / "primarily commercial contracts" / "mix of personal injury, family, and crime"}
Using the framework below, produce a pre-adoption compliance assessment:
1. **Data handling and retention**
- Where is data processed and stored (jurisdiction)?
- What is the provider's data retention policy (duration, conditions for extended retention)?
- Are prompts and outputs used for model training? Can this be contractually excluded?
- Does the provider offer zero data retention (ZDR) or equivalent?
- For cloud-hosted tools: who has access to the data (provider staff, subprocessors, law enforcement)?
2. **Confidentiality and privilege**
- Does uploading client material to this tool risk waiving legal professional privilege?
- What contractual commitments does the provider make on confidentiality?
- Does the tool's architecture allow data from one client matter to influence outputs for another?
- If the tool retains data, could that data be disclosed under a court order or regulatory request?
3. **Regulatory compliance**
- Does the tool's use comply with the SRA Standards and Regulations (or your regulator's equivalent), in particular:
* Duty of competence (can fee earners using this tool verify its outputs?)
* Confidentiality and disclosure obligations
* Supervision requirements (who is responsible for AI-generated work product?)
- Does the tool's use trigger obligations under UK GDPR or the Data Protection Act 2018?
- Does it involve automated decision-making within the meaning of Article 22 UK GDPR?
- Are there any sector-specific regulatory requirements (e.g., FCA, PRA) that apply to your clients' data?
4. **Supervision and verification**
- What internal processes will your firm put in place to verify AI outputs before they are used?
- Who is accountable for AI-generated work product at each level (fee earner, supervisor, partner)?
- How will AI use be documented and disclosed (to clients, to the court, internally)?
5. **Vendor risk**
- What happens to your data if the provider is acquired, ceases trading, or changes its terms?
- Does the contract include adequate termination, data return, and deletion provisions?
- Is the provider subject to foreign laws (e.g., US CLOUD Act, FISA 702) that could compel disclosure of your data?
6. **Recommendation**
Based on the above, classify this tool as:
- GREEN: suitable for adoption with standard controls
- AMBER: suitable for adoption with additional safeguards (specify which)
- RED: not suitable for adoption without significant changes to terms or deployment model
Constraints:
- {Add any firm-specific constraints, e.g., "all data must remain in the UK or EEA" / "we cannot accept any data retention beyond the session" / "we need to be able to demonstrate compliance to the SRA on request"}
- Apply the regulatory framework of England and Wales throughout
- Do not invent product features, contractual terms, or regulatory guidance
- Where information about the product's data practices is not publicly available, flag it as a gap and recommend what to ask the vendor
- This is a compliance planning tool, not legal advice
How did we do?
Hit reply and tell me what you would like covered in future issues or any feedback. We 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!
