This week a private key behind web encryption turned out to be the number 4. PHP2Shell knocked 32 UK government services offline. Railway told every user to rotate every secret.
Engineers are patching. For lawyers, the work is just starting.
What each side keeps missing 🧵
@lawdotcom mergers fix balance sheets faster than they fix partner incentives. the combinations that hold usually settle comp, conflicts and the tech stack before the announcement. otherwise the rainmakers just compare offers at the bigger firm
@NxtPrecedentPod the tell is usually the billing model. if a workflow change cuts hours and nobody has decided who keeps that margin, partners quietly route work back to the old way. tech plans that skip pricing and comp tend to stall at the pilot stage
Harvey becoming Wimbledon's official legal assistant partner says a lot about where legal AI is heading. the buyer isn't only the GC now, it's the brand team too.
when your in-house legal tool is also a sponsor, who audits the vendor you're contractually meant to promote?
@legalfutures £60k is the small number here. the bigger bill is client notifications, the insurer conversation at renewal and the panel reviews that follow. for a law firm a breach is a regulatory problem and a business development problem at the same time
@Shaileshwar21 This is why legal research has evolved from 'finding the law' to 'choosing your own adventure.' Conflicting precedents are officially the new normal 😁
@EamonLorincz the scoping letter does most of the work here. rule 1.2(c) allows limited scope if it's reasonable and the client gives informed consent, so spell out what the AI-assisted pass covers, what you personally reviewed, and what they're still on their own for
@RunOnNorth on the 'are people actually using it' part: logins and prompt counts flatter adoption. the better signal is how much AI-drafted text survives into what goes to the client or the court. that number tells you whether lawyers trust it or just open it
The ECtHR just barred a lawyer for two years over AI-drafted filings with invented cases and fake paragraphs from real judgments.
The fake paragraphs are the scarier part. A citation check passes them because the case exists. Verify the quoted text, not just the cite.
@gmdickinson the stopword list is doing the damage. 'and', 'or', 'not', 'any', 'no' all get stripped as noise, and in drafting those words often carry the entire operative meaning. legal NLP should start with an empty stopword list and make every removal earn its place
@TheLegalStack a sports body's in-house team is an interesting test case. the work is spiky, sponsorship, broadcast and player issues all landing in the same few weeks. if the tool holds up during the championship crunch, that says more than most law firm pilots.
an AI "court" where several LLMs vote on disputes between autonomous agents is a clever idea. the hard part isn't the verdict though, it's enforcement and appeal. if neither side can take the outcome to a real court, it's arbitration without a seat.
@gabepereyra the overlap that matters most is version control thinking. good engineers don't trust a change they can't diff. lawyers who treat every redline and prompt like a commit with a reason attached end up with a matter file that explains itself.
Every lawyer has a workflow they quietly fixed for themselves. A checklist, a template, a prompt that actually works. Most of it never leaves their desk. The profession would move faster if more of that got shared between peers.
@kkpeckk small law firms look the same from where i sit. a 10-lawyer practice won't spend weeks configuring a general tool. they'll pay for something that already knows their forms, their court's filing rules and their billing codes. setup cost is the real barrier, not price.
@WeAreLegora certs answer whether the vendor is safe. partners on a live matter ask something narrower: which documents did the agent actually touch, and what left the building? a per-matter audit log they can open themselves will probably earn more trust than any badge.
@Javed_Qadruddin the hard part is sorting which of the five are about the tool and which are just how that lawyer likes to work. two partners at the same firm will want the same memo formatted differently, so shipping every suggestion gets you a product built for one person
@filipoviclazar_ for a legal buyer the hero probably needs one thing more than speed: every flag tied to the exact page in the record. a PI lawyer can't put an AI summary in front of an expert or opposing counsel unless they can show where each fact came from
A literary prize just cleared a writer after AI detectors flagged his winning story.
A detector score is a probability, not proof. Once schools, employers and courts start treating it as evidence, the real fight will be over who carries the burden of disproving it.