In partnership with
The Law Commission published proposals on Thursday which would bring software and AI systems within the product liability regime for the first time, and that is the lead this week because it changes the advice many firms will be giving to clients who build or sell anything with AI inside it. Elsewhere, the incoming Master of the Rolls wants a page limit for litigants in person who use AI to state their case, the Law Society has published research warning that agentic AI is arriving in the justice system before anyone has decided where it should stop, and Legora has released a library of written instructions for its legal agents which firms can read and edit.
AI in Practice
AI systems are about to be treated as products, with strict liability attached
On 8 October the Law Commission opened a consultation on reforming the product liability regime in the Consumer Protection Act 1987, which imposes liability without proof of fault on producers of defective products which cause damage. The central proposal is a "broader, future-proof definition of 'product'" covering software and AI systems, whether they are embedded in a physical item or supplied digitally, together with the design features of social media and other online platforms. The paper also proposes extending liability to online marketplaces, clarifying when compensation is available for psychiatric illness, and introducing a mechanism to help claimants who face exceptional difficulty proving that a product was defective. The existing defences are to be retained, "subject to updates to ensure they continue to work effectively in cases concerning digital products". Professor Solène Rowan, quoted in the Law Commission's announcement, put the rationale simply: "The law must evolve with technology." The consultation closes on 14 January 2027, with final recommendations expected in late 2027.
The Act is nearly 40 years old and was written for physical goods, so whether software counts as a "product" has been an open question for most of the period in which software has mattered. The EU has already revised its own product liability rules to bring software within them, and the Law Commission's proposal is broadly similar. For anyone advising a business which develops AI, integrates a third party model into something it sells, or distributes AI-enabled products to consumers, the practical consequence is that a claimant injured by a defective AI system may in future not need to prove negligence at all. Liability under the Act cannot be excluded or limited by contract as against the person who suffers damage.
This author would add a note of caution in the other direction. The regime as it stands compensates for death, personal injury and damage to private property, and there is nothing in the published summaries to suggest that it will extend to pure economic loss. A law firm whose legal AI tool produces a wrong answer, which then finds its way into advice, should not expect these reforms to give it a claim against the vendor. For that, the firm's own contract with the vendor remains the only protection, and many firms will have accepted standard terms with tight liability caps. The detail on how the defences (in particular the defence that the state of scientific and technical knowledge could not have revealed the defect) will apply to systems which change after supply through updates or continued learning is the part this author will be reading most closely, because it is where the practical exposure of AI developers will be decided. These are proposals only, and legislation is some way off.
Read: Law Commission, and the reporting at Legal IT Insider
On your radar
The incoming Master of the Rolls wants a page limit for litigants in person: Sir Colin Birss, who became president of the new Business & Property Division on 1 October and takes over as Master of the Rolls on 1 November, told the International Conference of Legal Regulators in London last week that a strict limit of around five pages for a litigant in person to set out their case (with no limit on exhibits) would help judges cope with AI-drafted claims, adding: "I think there's something to be done there, and I propose to do it." He was even-handed about the technology, saying that material from litigants in person using AI often explains their case more clearly than it otherwise would, and that for many of them "it is better than what they had before". He supported the Civil Justice Council's work on transparency about AI use in expert evidence, while suggesting that for other documents lawyers' professional responsibility should govern disclosure. Why it matters for UK lawyers: this follows the EAT's judgment in Hancox, covered here on 25 September, and suggests the civil courts will deal with AI-generated volume through procedural limits rather than prohibition. If you act against unrepresented parties, keep a note of where an opponent's documents have caused disproportionate cost, because that is the evidence which will shape any new rule. (Legal Futures)
The Law Society says agentic AI may reach the courts before the rules do: A foresight report published on 6 October, The future of agentic AI in the justice system, drawn from interviews with practitioners, technologists, judges, academics and policymakers, warns that agents which act on users' behalf could become a "default doorway" into the justice system, and that the technology "will not fix a failing justice system" but risks reinforcing existing weaknesses such as underfunding and poor data. It calls for clear accountability, transparency of AI decision-making, meaningful human oversight, and public debate before widespread use. Vice president Brett Dixon's summary is the line most firms could adopt as policy: "Legal judgement cannot be delegated." Why it matters for UK lawyers: the report is about courts and tribunals, but its concern about where human judgement stops applies equally to any firm deploying agents on client work. Circulate it to whoever owns your AI policy and ask whether the policy says which decisions an agent may never take. (Law Society, Law Society Gazette)
Legora's agents now run on written instructions which firms can read and edit: Legora launched Skills on 7 October, a library of more than 250 sets of instructions written by its legal engineers across practice areas and jurisdictions, which tell its agents how to approach particular tasks. Each Skill is, in Legora's words, "a set of natural language instructions, saved as a Markdown file", and firms can read, edit and export them, adding their own templates, past examples and examples of what not to do. The announcement does not say how many of the Skills are written for English law. Why it matters for UK lawyers: an instruction file which governs how an agent does the work is, in effect, a precedent or a procedure, and it needs an owner and a review date in the same way. If your firm uses Legora, find out who is allowed to edit Skills, check whether any you rely on were written for another jurisdiction, and keep a record of which version was in use on a given matter. (Legora, Artificial Lawyer)
The new CoCounsel Legal is on its way to UK firms: Thomson Reuters made the next generation of CoCounsel Legal generally available in the US on 20 August, describing it as a fully agentic product built on Anthropic's Claude Agent SDK, grounded in Westlaw and Practical Law, and including a Brief Builder which takes a litigator from research to a first draft. The UK, Canada and Australia were promised "later this year". This author has had a demonstration of the new version and has booked a trial. We will report properly once trialled. Why it matters for UK lawyers: Thomson Reuters says the product works "at the level of a senior associate", which is a vendor claim rather than a finding, and the checking duty set out in Hancox applies to a Brief Builder draft as much as to any other. If your firm is offered the upgrade, ask for a trial on English authorities you already know well before agreeing to new terms. (Thomson Reuters)
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.
Speak naturally. Send without fixing.
Wispr Flow turns your voice into clean, professional text you can send the moment you stop talking. No rough transcription you have to clean up. Actual polished text, ready for email, Slack, or any app.
Speak the way you think. Go on tangents. Change your mind mid-sentence. Flow strips the filler, fixes the grammar, and gives you text that reads like you spent five minutes writing it.
For Review
The future of agentic AI in the justice system (Law Society)
The full foresight report behind the second radar item. Read the sections on accountability and on how "routine" work is defined, because the warning that delegation can happen gradually, as cases which look routine are handed to systems one at a time, applies to a firm's own processes as much as to the courts.
Read or listen: Law Society
Private client roundtable: "It feels like it's devaluing our profession" (Law Society Gazette)
Not this week's news: published on 22 September. Private client practitioners describe clients arriving with AI research which is wrong, complaints drafted by ChatGPT citing law which does not exist, and the unbilled time spent correcting both. One firm now states in its client care letter that emails are charged in units, which is a practical answer worth considering if your own fee earners are absorbing the same cost.
Read or listen: Law Society Gazette
Can agentic AI be controlled? (New Law Journal)
A short piece from 7 October drawing on the Law Society report, which is worth reading for its point that agentic systems assemble materials, identify authorities and organise reasoning in a way which makes it hard to see where the human decision began. That is the supervision problem in a sentence, and a useful framing for any internal training session on agents.
Read or listen: New Law Journal
Practice Prompt
Try the below prompt to map a client's exposure if software and AI systems are brought within the product liability regime, as this week's Law Commission consultation proposes. 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 assisting a solicitor in England and Wales advising a business client which develops, integrates, distributes or sells products containing software or AI systems. Your task is to produce a first-pass exposure map showing where the client could sit in the supply chain if the Law Commission's October 2026 proposals to bring software and AI systems within the Consumer Protection Act 1987 are implemented. You are not predicting the final law and you are not advising on the merits of any claim.
Context to apply:
- The client's business and its products, described without names or identifying detail: {e.g. a UK company which builds a third party AI model into a consumer home device sold through its own website and an online marketplace}
- The AI or software components and who supplies each one: {e.g. foundation model licensed from a US provider, firmware written in-house, mobile app built by a contractor}
- How the product reaches consumers: {direct sale, online marketplace, distributor, fulfilment service, importer}
- How the software or AI changes after supply: {e.g. automatic updates, continued learning, none}
- The relevant contracts, summarised: {supplier licences, distribution agreements, marketplace terms, with their indemnity, liability cap and insurance clauses}
- The client's insurance cover for product liability: {summarise, or say "not yet checked"}
- Any known incidents, complaints or defect reports: {describe, or say "none"}
Produce the map under these headings:
1. Where the client sits
Identify each role the client could occupy under the current Act (producer, own-brander, importer, supplier) and under the proposals as publicly summarised (including any new categories such as online marketplaces or fulfilment service providers). Distinguish clearly between the current law and what is only proposed.
2. Upstream and downstream parties
For each other party in the supply chain, state the role it would be likely to occupy and whether the client could recover from it, or be pursued by it, under the contracts as summarised.
3. Contract gaps
Identify each contract term which appears to assume that software or AI is not a product, each indemnity which does not reach product liability, and each liability cap which would leave the client carrying a loss it cannot pass on. Remember that liability under the Act cannot be excluded or limited as against the person who suffers damage.
4. Updates and learning
Set out the questions the client must answer about who controls updates and changes to the AI after supply, and why those answers may matter to whether a defect existed at the relevant time and to any available defence.
5. Insurance
List the questions to put to the client's broker about whether current cover would respond to a strict liability claim involving software or AI.
6. Consultation response
Identify any points on which the client has a commercial interest in responding to the consultation before it closes on 14 January 2027, and what evidence it would need to make the point well.
7. What this map cannot tell you
Flag every point resting on an assumption about the proposals, the contracts or the facts, and say what would need to be established to replace it with a fact.
Constraints:
- {Add matter-specific constraints, for example a product launch date, an existing claim, a client instruction to focus on one market, or an insurer with notification requirements.}
- Apply the law of England and Wales. Treat the Law Commission's proposals as proposals only, and where you are unsure what the consultation paper says on a point, say so and tell the solicitor to check the paper itself.
- Do not invent statutory provisions, case law, consultation questions or contract terms. If you cannot verify something, mark it as unverified.
- Do not conclude that the client is or is not liable, or that any contract is adequate. The purpose is to identify exposure and open questions.
- Write in plain English, and keep each finding short enough to act on.
- This is a planning aid, not legal advice. The solicitor remains responsible for the advice given to the client.How did we do?
Hit reply and tell me whether your clients have started asking about liability for the AI in their own products.
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!
