Jiten’s Jobs exists because TCAP wanted a closer look at the institutions that helped form Jiten Kotecha, the senior employment lawyer now occupying what we regard as a deeply dishonourable role inside Cummins. Pinsent Masons was where his legal career began, first as a trainee and later as an employment lawyer. Kotecha was long gone by 2026, but his old training ground has since produced a legal-tech episode too perfect to leave alone. Pinsent Masons publicly warned lawyers about the dangers of trusting artificial intelligence without checking its work. Months later, invented statutory wording reached the High Court on Pinsent Masons letterhead, the court was misled twice, the firm referred itself to the regulator and, twenty-six days after judgment, Pinsent launched a generative-AI employment tribunal service built around lawyer supervision.
There is an easy version of this story where everybody blames the robot. Artificial intelligence hallucinated. Artificial intelligence invented law. And it confidently produced something that looked authoritative while quietly having no fucking idea whether it was true. All of that happened, but it misses the more interesting point. The AI was not a solicitor, did not owe duties to the court, did not supervise anybody and did not decide to send anything on Pinsent Masons letterhead. The machine supplied the bullshit. The lawyers supplied the professional authority.
That is the thesis here. This was not primarily a technology failure. It was a supervision failure inside a major international law firm that had already publicly explained, in perfectly clear English, how the technology was supposed to be controlled. Pinsent Masons knew the risk, published the warning, wrote internal safeguards and still managed to let fictional law travel through human review and into the High Court.
Welcome back to Jiten’s Jobs.
Pinsent Masons Had Already Written The Warning Label
On 17 June 2025, Pinsent Masons published an article about the dangers of lawyers relying on artificial intelligence without independently verifying what it produced. The firm was commenting on High Court proceedings in which false authorities and unreliable AI-assisted research had reached the court, and its message was not buried beneath twelve pages of consultancy fog. AI might help with research, but “outputs must be checked for accuracy” and lawyers supervising junior staff had to ensure that AI-assisted work was independently verified against authoritative legal sources.
That was sensible advice because generative AI has a particular talent for producing elegant nonsense. Ask it a legal question and it can return case names, statutory provisions and polished reasoning with the confidence of something carved above the doors of the Supreme Court. The problem is that confidence is presentation, not proof. Sometimes the case does not exist. Sometimes the quotation is wrong. Then sometimes the machine finds exactly the law you hoped existed because it has simply invented the fucking thing.
Pinsent Masons understood that problem well enough to explain it to the rest of the profession. Its own internal AI policy also warned that outputs could be inaccurate or fabricated, required human supervision and made users responsible for applying critical thought before relying on anything the technology produced. The firm therefore entered 2026 with the warning signs already installed, the risk already identified and the checking procedure already written down.
Then came Cork v Smith.
A Boring Insolvency Application Develops A Plot
The case concerned a block transfer application in the High Court’s Insolvency and Companies List. These applications are normally administrative machinery, used where insolvency practitioners need appointments transferred in bulk. Nobody arrives expecting fireworks. Nobody commissions a Netflix adaptation. The paperwork goes in, the court deals with the necessary powers and everybody moves on with their lives.
This particular application also sought the release of an outgoing liquidator from liability. ICC Judge Mullen wanted to know where the court’s power to grant that release came from, so on 20 March 2026 the court asked Pinsent Masons a straightforward legal question. There was an express power concerning administrators, but the judge could not see the equivalent power covering the liquidator in the circumstances before him.
This was exactly the sort of question for which people employ a large commercial law firm. Find the relevant legislation, analyse it, explain the answer and tell the court where the power sits. The entire fucking profession is built around turning questions like that into reliable answers.
Pinsent Masons turned to artificial intelligence.
The Machine Finds Exactly The Law Everyone Needed
A junior associate, anonymised in the judgment as Lawyer A, used an AI tool being piloted by Pinsent Masons to research the issue. According to the judgment, the hallucinations began almost immediately, with the system producing unreliable material while working its way through the insolvency legislation. Eventually it supplied purported wording attributed to Rule 12.37(5) of the Insolvency (England and Wales) Rules 2016.
The wording was tremendously convenient because it appeared to give the court the very sort of power Pinsent Masons was trying to establish. It looked like statutory language, sounded like statutory language and solved the statutory problem. If you were designing a piece of legislation specifically to make the firm’s life easier that afternoon, you might have produced something remarkably similar.
Unfortunately, Parliament had neglected to write it.
The supposed wording was not contained in Rule 12.37(5). It was not sitting elsewhere in Rule 12.37 under the wrong paragraph number, nor could Judge Mullen locate it anywhere else in the Insolvency Rules. The actual provision dealt with who could apply where an outgoing office-holder was an administrator. The useful wording supplied by the AI simply did not exist.
The technology had not discovered an obscure rule.
It had written one.
The Robot Was Apparently More Cautious Than The Humans
The episode becomes much harder to dismiss as an unavoidable technological accident because the AI itself repeatedly indicated that its output needed verification. Judge Mullen’s examination of the chat transcripts showed that the program warned about checking the legislation and using authoritative sources before relying upon what it had produced.
That should have been the easiest part of the exercise. Nobody needed to excavate an archive beneath Lincoln’s Inn or telephone a retired parliamentary draftsman living in Dorset. The legislation was available. Open the source, find Rule 12.37(5), compare the words on the screen and discover that the helpful passage generated by the machine was not fucking there.
That verification did not happen. Judge Mullen later said Lawyer A appeared to have “almost entirely outsourced the thinking process” to the AI program. It is difficult to improve upon that description because it captures the fundamental problem more neatly than any lecture about legal technology could. The tool had ceased to be an assistant and had become the person in the room apparently doing the thinking.
The output then entered the supervision chain.
Good To Go
The matter was being handled day to day by Lawyer A and senior associate Samantha Poulton, with partner Steven Cottee supervising. The draft correspondence containing the purported statutory wording passed through senior review before it went anywhere near the court.
Cottee approved the material as “good to go” without checking the supposed statutory wording against the actual legislation. Poulton subsequently approved the correspondence as “fine”, having understood that the partner had already reviewed it. The junior had trusted the AI, the partner had taken the material at face value and the senior associate had relied on the partner’s review.
This is where the story stops being about a junior lawyer pressing the wrong button. Law firms charge for layers of professional judgment precisely because less experienced lawyers will sometimes get things wrong. Research is reviewed, drafts move upwards and senior practitioners are supposed to notice when the central legal proposition answering a judge’s direct question has apparently fallen out of the sky wearing a rule number.
Pinsent Masons had the layers. What it did not have, at the moment that mattered, was somebody opening the fucking legislation.
The letter went out.
The Judge Checks The Lawyers’ Homework
Pinsent Masons’ letter of 30 March 2026 identified Rule 12.37(5) as containing an express power and presented the invented wording in a form that looked like a quotation from the legislation. The passage was set apart and styled as statutory text, giving the court every reason to understand that Pinsent Masons was telling it what the rule actually said.
Judge Mullen did something surprisingly unfashionable.
He checked.
The words were not there. He looked elsewhere in the relevant chapter in case the paragraph number was merely wrong, then searched the Insolvency Rules for phrases from the purported provision. Still nothing. His immediate suspicion was that Pinsent Masons had put an AI hallucination before the court without verifying it.
There is something gloriously upside-down about the scene. The judge had asked the solicitors for the law, the solicitors had supplied something that looked like law, and then the judge had to conduct his own research to establish whether the law supplied by one of Britain’s major commercial firms existed in the first place.
Premium legal services had developed a fact-checking stage.
The customer was the High Court.
The Court Gives Them A Second Chance
On 14 April 2026, the court confronted Pinsent Masons directly. Judge Mullen explained that Rule 12.37(5) did not contain the words the firm had attributed to it and that he could not find them elsewhere in the rules. Pinsent Masons was asked to identify where the wording came from and explain why it had told the court that an express statutory power existed when the cited provision contained no such thing.
This was the moment to stop digging. The firm could have gone back through the research trail, identified the AI hallucination, admitted what had happened, corrected the legal position and apologised. The first letter would still have been embarrassing, but professional life contains embarrassing mistakes. The important thing is what happens when somebody discovers one.
Instead, the second response attempted to explain the earlier passage as effectively a summary of the legal position rather than what it plainly appeared to be, namely purported statutory wording from Rule 12.37(5). Judge Mullen rejected that explanation as “not credible”. The first letter had specifically referred to the paragraph, called the power express and introduced text styled like a quotation from the rule.
The second letter had also been produced with AI assistance.
The imaginary law now had an imaginary explanation.
Misled Twice
By this point, the damage was no longer confined to an incorrect legal proposition. The judge had begun questioning whether he could rely upon other material Pinsent Masons had placed before the court, which transformed what should have been a routine administrative application into a hearing about the solicitors themselves.
Witness statements were required explaining how the letters had been produced. The lawyers involved had to account for the process, the firm’s AI conversations were disclosed and another firm, Irwin Mitchell, took over representation of the applicants. Pinsent Masons agreed to meet the additional costs caused to its former clients by the episode, including those arising from the need for replacement solicitors.
Judge Mullen recorded that a substantial amount of court time had already been wasted dealing with something that should never have required judicial attention. Artificial intelligence is usually sold to the legal industry on efficiency, speed and the elimination of repetitive work. Pinsent Masons had found a more innovative application.
It used AI to manufacture an extra court hearing.
The judge’s summary was brutal enough without TCAP adding anything to it. The court had been “misled not once but twice”.
The Policy Said Exactly What Everybody Should Have Done
The most damaging part of this story is not that Pinsent Masons lacked controls. It had them.
Its AI policy warned that outputs might be inaccurate, required human supervision and specifically cautioned that apparently believable material might be outdated, misleading or completely fabricated. Users remained responsible for checking the output and applying critical thought before relying upon it.
That is almost a complete description of what did not happen in Cork v Smith.
The policy knew artificial intelligence could hallucinate. The firm’s public commentary knew artificial intelligence could hallucinate. The AI itself reportedly warned that its legal material needed checking. Yet the invented wording travelled through the junior lawyer, through senior review and eventually into correspondence with the court.
There is a limit to what another policy can achieve at that point. You can laminate the warning. Put it beside every monitor. Print it on mugs. Have somebody in Risk Management perform it as interpretive dance at the Christmas party.
Eventually somebody still has to check the fucking rule.
Public Admonishment
Pinsent Masons referred itself to the Solicitors Regulation Authority and told the court that it would cooperate with any investigation. The firm also apologised, supplied the relevant AI material and described further safeguards it was putting in place. Judge Mullen expressly recognised that Pinsent Masons was taking the matter seriously once the full problem emerged.
He nevertheless concluded that publication of the judgment would serve as a public admonishment of Pinsent Masons, Cottee and Poulton, alongside referral to the regulator. The judge considered there to be at least a prima facie case concerning professional duties not to mislead the court and not to waste court time.
This was not a finding that the senior lawyers had deliberately conspired to deceive anybody. The uglier lesson was more ordinary than that. Experienced professionals working inside a reputable law firm had allowed unreliable AI-assisted work to acquire institutional authority because each layer of supervision assumed enough checking had happened somewhere else.
Nobody had to be plotting anything.
Everybody merely had to trust the person below them.
That was sufficient.
The Machine Did Not Practise Law
Artificial intelligence deserves criticism for hallucinating legal material, but treating the technology as the principal culprit lets the humans off far too cheaply. The machine did not owe a professional duty to Judge Mullen, did not sign correspondence, did not supervise Lawyer A and did not decide that the draft was ready to leave the building.
The lawyers did.
That matters enormously as firms race to incorporate generative AI into legal services. The selling proposition is that machines can process routine work more quickly, identify patterns across enormous datasets and leave expensive human lawyers free to concentrate on judgment, strategy and the difficult bits.
Fine.
But if the expensive human layer becomes a ceremonial rubber stamp, the whole fucking proposition collapses. You have not automated legal work. You have automated the production of material wearing a law firm’s name.
In Cork v Smith, the AI supplied the fiction.
Pinsent Masons supplied the authority.
Twenty-Six Days Later
Judge Mullen handed down judgment on 22 May 2026.
On 17 June 2026, twenty-six days later, Pinsent Masons announced an enhanced employment tribunal service incorporating proprietary generative-AI workflows.
It was called Employment Tribunals Reimagined.
There is almost no editorial work required here.
Pinsent said the service combined its employment-law expertise with AI workflows and managed legal services to help employers process growing volumes of tribunal claims. Repeatable parts of the claims process could be automated, cases could be triaged by complexity and Pinsent’s lawyers could focus their time on areas that added the greatest value.
Most importantly, the firm explicitly said those AI workflows would always be supervised by lawyers.
Twenty-six days after a High Court judgment dissecting precisely what can happen when lawyer supervision becomes assumption rather than action, the word supervised had acquired one hell of a fucking footnote.
Exactly One Year Of Legal Technology
The dates produce an almost absurdly neat triptych.
On 17 June 2025, Pinsent Masons publicly warned that AI-assisted legal research had to be verified and that supervising lawyers bore responsibility for ensuring junior work was accurate. On 22 May 2026, the High Court published a judgment explaining how invented legal material had nevertheless travelled through Pinsent Masons and misled the court twice. Then, on 17 June 2026, exactly one year after its public warning, Pinsent Masons launched a new employment tribunal proposition built around generative AI and lawyer supervision.
No metaphor improves that chronology.
It is already a fucking satire.
The firm had spent one June telling lawyers how not to use AI, the following spring demonstrating why, and the next June selling an AI-assisted legal service in the employment field where Jiten Kotecha himself had once worked.
The technology had completed a full lap.
AI Is Apparently Part Of The Problem And The Solution
Pinsent Masons’ launch material contained another wonderful detail. The firm said the employment tribunal system was under mounting pressure and pointed to employees increasingly using generative AI to produce formal grievances and tribunal claims, which could be lengthy, complicated and poorly structured. Employers were therefore facing additional correspondence and greater cost.
Then came Pinsent’s solution.
Generative AI.
There is nothing inherently incoherent about that. A controlled professional tool can obviously be used more effectively than an employee asking a public chatbot to write seventy-three pages about a disciplinary meeting. The underlying commercial logic makes sense.
The scene is simply too good not to enjoy.
Employees use AI to generate enormous grievances. Employers become buried beneath AI-generated material. Pinsent Masons deploys its own AI to help the employers process the employees’ AI.
Soon the Employment Tribunal can simply install two servers opposite each other and let them argue until the fucking preliminary hearing.
Jiten’s Old Training Ground
Jiten Kotecha started his legal career at Pinsent Masons and later worked in the firm’s employment practice. Pinsent’s own archives name him on the employment side of a corporate acquisition in 2013, placing his early career squarely inside the kind of sophisticated commercial environment that later fed into his senior in-house employment role at Cummins.
That is why this belongs in Jiten’s Jobs.
TCAP began looking backwards through Kotecha’s professional history because of the role he now occupies and the conduct around the Cummins employment file which we regard as dishonourable. The point was never simply to list employers from a LinkedIn profile. We wanted to inspect the institutions, cultures and professional environments behind the polished biography.
Pinsent Masons trained lawyers, advised major corporations and sold high-end professional judgment. It also understood the risks of artificial intelligence well enough to warn everybody else about them.
Then fictional law passed through its own supervision structure and landed before a High Court judge.
That was not Jiten Kotecha’s mistake.
It was his former professional habitat producing another remarkable exhibit.
The Rule That Didn’t Exist
Law firms are going to use more artificial intelligence. They should. Used properly, the technology can reduce repetitive work, accelerate analysis and make professional services cheaper and more efficient.
But Cork v Smith exposes the point where all the glossy legal-tech presentations eventually arrive.
Somebody still has to think.
Somebody still has to ask whether the answer makes sense, open the authoritative source, compare the wording and accept personal responsibility for what leaves the building. A sophisticated AI workflow cannot rescue a professional culture in which verification becomes something everybody assumes somebody else has already done.
Pinsent Masons knew that before March 2026. Its public guidance said so. Its internal policy said so. The AI itself apparently said so.
The invented rule still reached the High Court.
Twenty-six days after the judgment, Employment Tribunals Reimagined arrived.
One involved reimagining how employment litigation could be delivered.
The other involved reimagining what Parliament had fucking written.
Only one was supposed to be fictional.
Lee Thompson – Founder, The Cummins Accountability Project
Sources
- The National Archives – Anthony Malcolm Cork & Anor v Mark Smith [2026] EWHC 1199 (Ch)
- BAILII – Malcolm Cork & Anor v Smith [2026] EWHC 1199 (Ch)
- Pinsent Masons – English High Court Highlights That Lawyers Using AI Must Ensure Accuracy
- Pinsent Masons – Employment Tribunals Reimagined
- Law Gazette – Pinsent Masons And Solicitors Referred To SRA After AI Failures
- Legal Futures – Junior Solicitor Used AI To Draft Misleading Letters To Court
- Pinsent Masons – Pinsent Masons Advises Coaxis On The Purchase Of 4Projects
- Solicitors Regulation Authority – Jiten Kotecha