How Alexander Kardos-Nyheim turned a side project into a 'life-changing' exit while he trained at one of the world's top law firms
Alexander Kardos-Nyheim is the founder of Safe Sign Technologies, the legal tech startup he started as a side project during his training contract at A&O Shearman.
In August 2024, global business media giant Thomson Reuters acquired Safe Sign Technologies for a “life-changing” sum of money - just as Alexander was qualifying as a lawyer.
We asked Alexander to tell us more about the company he founded, what it was like to juggle a startup with the demands of a training contract at one of the world’s top law firms, and whether he ever needs to work again now he’s successfully exited the business.
This is his story…
Get the free email for UK lawyers with the legal industry and business stories you need to know about to stay ahead. In your inbox, three times a week.
The company and the product
Safe Sign Technologies is first and foremost a story of an unusual combination of brilliant scientists and lawyers who joined me - a 21-year-old law graduate - to try to achieve something of social and technological importance.
They put their trust in me, and everything on the line, to achieve a shared, and frankly mad, goal. And we pulled it off - more on that later.
Turning first to the company and the product: I founded Safe Sign Technologies in 2022 to create the world’s best Large Language Model for Law, with the mission of making the world’s best legal advice accessible to all.
The product is a specialised, robust, proprietary LLM that is the most capable AI system in the world on a huge range of legal tasks, beating multi-billion dollar AI giants like OpenAI.
Much of how we did this remains a secret, but I’ll try to disclose as much as I can in the hope that our story can encourage others to pursue their ambitions, no matter how far-flung they may seem.
We had two problems in mind…
Problem 1: The tech problem
We all know OpenAI’s GPT models, and many of us know the emerging rivals: Anthropic’s Claude models, Meta’s Lama models, Google DeepMind’s Gemini models, etc.
What these companies have in common is that they are creating Large Language Models that can do a very broad range of tasks to differing levels of success. The focus is on breadth, not depth.
These companies are trying to capture the next billion customers - to help 14-year-olds with their English homework, to generate quick PowerPoint presentations for overworked junior office workers, or to speed up report writing for executives.
These models are impressive at being generalists, but not specialists: the moment you need anything beyond the superficial, you will be disappointed.
Pushing these models to give accurate, reliable details on specialist areas like law, medicine and engineering leads to very high levels of hallucinations (fabricated, often nonsensical outputs) which serious people cannot rely on, and which risk misleading professionals into giving negligent advice.
This isn’t surprising: these models are trained on the good, the bad and the ugly of the World Wide Web, with little filtering.
Human training is typically limited to poorly paid workers in third-world countries, the data these models are fed with are often illegally or unethically sourced (see, for instance, the New York Times’ suit against OpenAI), and the scientific ethos of these companies - whatever some of them may say - is to prioritise speed of development over safety, robustness and reliability.
Release first, ask questions later.
That’s fine for cheating on your English homework (though I would still trust myself more…): it’s plainly inadequate for the legal domain.
Whether you’re a lawyer, or an individual or company looking for legal information, there is no room for error.
Law is an accuracy-critical vertical where the stakes are always high. The legal arena is one where one’s rights and freedoms can be won or lost by the quality of one’s legal advice: the cost of a mistake can be existential.
If the domains of human pursuit could be categorised as types of surgery, I’d classify law as open-heart surgery.
I, for one, wouldn’t want a heart surgeon who received their training from Wikipedia and Reddit, and who is constantly trying to multi-task during the operation with stamp collecting, and who starts hallucinating giraffes as he cuts into my left ventricle.
I would want a heart surgeon trained hard and unrelentingly by the world’s best heart specialists; a surgeon tested on the most challenging cases, clear on the importance of their duty to the patient, and focused solely on the task of delivering a world-class surgical outcome.
So the fundamental tech problem we were trying to solve was that there is no serious AI answer to the legal domain: lawyers do not have the accurate, safe and reliable system they deserve in order to be expected to move into the AI age.
And the same things that made AI unreliable for lawyers were universal AI problems that hold the technology back from delivering some of its most impactful uses on human civilisation.
The fundamental tech problem we were trying to solve was that there is no serious AI answer to the legal domain: lawyers do not have the system they deserve to move into the AI age.
Problem 2: The access to justice problem
I spent my teenage years fighting to prevent a swathe of south-east London’s final working class community, including my mother’s house, from being demolished by a crude and aggressive luxury development project.
My mother, who had brought me up as a single parent and struggling artist, had only one asset amid all the difficulty she had faced since escaping Communism in the 70s: her apartment, which she relied on for her painting career, and to bring me up. Neither she, nor our community, could afford legal advice.
The roof over our head was literally at stake, and faced with a Magic Circle-advised property developer, the odds were stacked against us.
I was 14 at the time. I ended up having to fight the case on my own, representing my mum, metal workers, carpenters and their struggling families in front of several government committees.
I had to juggle my GCSEs with legal and historical research in archives and libraries to save these modest buildings that were of huge importance to their inhabitants.
Uncovering possible corruption and errors in the developers’ and government’s reasoning, I won the case. A local newspaper described it as a “David and Goliath” story.
But the fact that an entire community relied on one boy to save them from destitution and the loss of fundamental property rights revealed a deep access to justice problem.
Equal rights under the law relies on being able to enforce and protect those rights: without equal access to legal advice, we will never be an equal society. The David and Goliath metaphor shouldn’t apply.
Aged 15, I submitted law reform proposals to the Mayor of London and almost got a Private Member’s Bill moving in the House of Lords - but could not get the momentum to change the law that put everyday people at such risk.
I set up the Urban Development Reform Organisation which today fights for communities across the UK to level the playing field in the face of monolithic development projects, but this was only part of the solution.
I then thought: perhaps there is a technological angle to the access to justice problem.
This problem is ultimately a question of scale. Everyday people cannot easily access high quality legal advice because of its scarcity and expense. The advent of LLMs offered an opportunity to scale world-class legal advice to millions of people cheaply and instantly.
The solution
The solution was to develop the world’s best Legal Large Language Model, trained by world-leading lawyers and AI experts.
A model that surpasses Big Tech drawing on, and pioneering, the latest research on safety, robustness and reliability, so that lawyers and those seeking legal advice can actually trust the model’s outputs to deliver advice and plan their lives.
A model that does not rely on external AI model providers but retains strategic independence so that we can control pricing and the model’s wider architecture.
We aimed at nothing less than a profound reshaping of society into a model where expert legal advice is no longer a privilege but as ubiquitous to everyday life as the world’s collective knowledge has become through search on the internet.
I spent my teenage years fighting to prevent my mother’s house from being demolished by a crude and aggressive luxury development project.
Law Firm | Trainee First Year | Trainee Second Year | Newly Qualified (NQ) |
|---|---|---|---|
| Addleshaw Goddard | £52,000 | £56,000 | £100,000 |
| Akin Gump | £60,000 | £65,000 | £174,418 |
| A&O Shearman | £56,000 | £61,000 | £150,000 |
| Ashurst | £52,000 | £57,000 | £125,000 |
| Baker McKenzie | £56,000 | £61,000 | £140,000 |
| Bird & Bird | £47,000 | £52,000 | £98,000 |
| Bristows | £46,000 | £50,000 | £88,000 |
| Bryan Cave Leighton Paisner | £50,000 | £55,000 | £105,000 |
| Burges Salmon | £47,000 | £49,000 | £72,000 |
| Charles Russell Speechlys | £50,000 | £53,000 | £88,000 |
| Cleary Gottlieb Steen & Hamilton | £57,500 | £62,500 | £164,500 |
| Clifford Chance | £56,000 | £61,000 | £150,000 |
| Clyde & Co | £47,000 | £49,500 | £85,000 |
| CMS | £50,000 | £55,000 | £110,000 |
| Cooley | £55,000 | £60,000 | £157,000 |
| Davis Polk & Wardwell | £65,000 | £70,000 | £170,000 |
| Debevoise & Plimpton | £55,000 | £60,000 | £168,000 |
| Dechert | £55,000 | £61,000 | £165,000 |
| Dentons | £50,000 | £54,000 | £100,000 |
| DLA Piper | £50,000 | £55,000 | £110,000 |
| Eversheds Sutherland | £46,000 | £50,000 | £100,000 |
| Farrer & Co | £47,000 | £49,000 | £88,000 |
| Fieldfisher | £48,500 | £52,000 | £95,000 |
| Freshfields | £56,000 | £61,000 | £150,000 |
| Fried Frank | £55,000 | £60,000 | £175,000 |
| Gibson Dunn | £60,000 | £65,000 | £180,000 |
| Goodwin Procter | £55,000 | £60,000 | £175,000 |
| Gowling WLG | £48,500 | £53,500 | £98,000 |
| Herbert Smith Freehills | £56,000 | £61,000 | £135,000 |
| HFW | £50,000 | £54,000 | £100,000 |
| Hill Dickinson | £43,000 | £45,000 | £80,000 |
| Hogan Lovells | £56,000 | £61,000 | £135,000 |
| Irwin Mitchell | £43,000 | £45,000 | £76,000 |
| Jones Day | £56,000 | £65,000 | £160,000 |
| K&L Gates | £50,000 | £55,000 | £115,000 |
| Kennedys | £43,000 | £46,000 | £85,000 |
| King & Spalding | £55,000 | £60,000 | £165,000 |
| Kirkland & Ellis | £60,000 | £65,000 | £174,418 |
| Latham & Watkins | £60,000 | £65,000 | £174,418 |
| Linklaters | £56,000 | £61,000 | £150,000 |
| Macfarlanes | £56,000 | £61,000 | £140,000 |
| Mayer Brown | £55,000 | £60,000 | £135,000 |
| McDermott Will & Emery | £65,000 | £70,000 | £174,418 |
| Milbank | £65,000 | £70,000 | £174,418 |
| Mills & Reeve | £45,000 | £47,000 | £82,000 |
| Mischon de Reya | £47,500 | £52,500 | £95,000 |
| Norton Rose Fulbright | £48,500 | £53,000 | £120,000 |
| Orrick | £55,000 | £60,000 | £140,000 |
| Osborne Clarke | £54,500 | £56,000 | £94,000 |
| Paul Hastings | £60,000 | £68,000 | £173,000 |
| Paul Weiss | n/a | n/a | £180,000 |
| Penningtons Manches Cooper | £48,000 | £50,000 | £83,000 |
| Pinsent Masons | £49,500 | £54,000 | £97,000 |
| Quinn Emanuel | n/a | n/a | £180,000 |
| Reed Smith | £50,000 | £55,000 | £125,000 |
| Ropes & Gray | £60,000 | £65,000 | £165,000 |
| RPC | £46,000 | £50,000 | £90,000 |
| Shoosmiths | £43,000 | £45,000 | £97,000 |
| Sidley Austin | £60,000 | £65,000 | £175,000 |
| Simmons & Simmons | £52,000 | £57,000 | £120,000 |
| Skadden | £58,000 | £63,000 | £173,000 |
| Slaughter and May | £56,000 | £61,000 | £150,000 |
| Squire Patton Boggs | £47,000 | £50,000 | £110,000 |
| Stephenson Harwood | £50,000 | £55,000 | £100,000 |
| Sullivan & Cromwell | £65,000 | £70,000 | £174,418 |
| Taylor Wessing | £50,000 | £55,000 | £115,000 |
| TLT | £44,000 | £47,500 | £82,000 |
| Travers Smith | £54,000 | £59,000 | £120,000 |
| Trowers & Hamlins | £45,000 | £49,000 | £80,000 |
| Vinson & Elkins | £60,000 | £65,000 | £173,077 |
| Watson Farley & Williams | £50,000 | £55,000 | £102,000 |
| Weightmans | £34,000 | £36,000 | £70,000 |
| Weil Gotshal & Manges | £60,000 | £65,000 | £170,000 |
| White & Case | £62,000 | £67,000 | £175,000 |
| Willkie Farr & Gallagher | £55,000 | £60,000 | £165,000 |
| Withers | £47,000 | £52,000 | £95,000 |
| Womble Bond Dickinson | £43,000 | £45,000 | £80,000 |
Rank | Law Firm | Revenue | Profit per Equity Partner (PEP) |
|---|---|---|---|
| 1 | DLA Piper* | £3,010,000,000 | £2,400,000 |
| 2 | Clifford Chance | £2,300,000,000 | £2,040,000 |
| 3 | A&O Shearman | £2,200,000,000 | £2,200,000 |
| 4 | Hogan Lovells | £2,150,000,000 | £2,200,000 |
| 5 | Freshfields | £2,140,000,000 | Not disclosed |
| 6 | Linklaters | £2,100,000,000 | £1,900,000 |
| 7 | Norton Rose Fulbright* | £1,800,000,000 | £1,100,000 |
| 8 | CMS** | £1,620,000,000 | Not disclosed |
| 9 | Herbert Smith Freehills | £1,300,000,000 | £1,300,000 |
| 10 | Ashurst | £961,000,000 | £1,300,000 |
| 11 | Clyde & Co | £844,000,000 | £739,000 |
| 12 | Eversheds Sutherland | £749,000,000 | £1,300,000 |
| 13 | BCLP* | £661,000,000 | £748,000 |
| 14 | Pinsent Masons | £649,000,000 | £793,000 |
| 15 | Slaughter and May*** | £625,000,000 | Not disclosed |
| 16 | Simmons & Simmons | £574,000,000 | £1,076,000 |
| 17 | Bird & Bird** | £545,000,000 | £696,000 |
| 18 | Addleshaw Goddard | £495,000,000 | Not disclosed |
| 19 | Taylor Wessing | £480,000,000 | £915,000*** |
| 20 | Osborne Clarke** | £456,000,000 | £771,000 |
| 21 | Womble Bond Dickinson | £448,000,000 | £556,000 |
| 22 | DWF | £435,000,000 | Not disclosed |
| 23 | Fieldfisher | £407,000,000 | £966,000 |
| 24 | Kennedys | £384,000,000 | Not disclosed |
| 25 | DAC Beachcroft | £325,000,000 | £700,000 |
What do City lawyers actually do each day?
For a closer look at the day-to-day of some of the most common types of lawyers working in corporate law firms, explore our lawyer job profiles:
I needed to assemble the world’s best team in the field, even though I had no business track record and, initially, no money.
I started by approaching my old Cambridge Law professors.
Professor Felix Steffek was the very first to join me: when I only had a half-pager capturing the idea, he decided to lend me his backing and expertise. As well as being a world-leader in Legal AI, he had previously written my exams, so you can imagine I had mixed feelings about him before I first met him.
Professor Jodi Gardner, who had taught me Tort Law - and who I presumed would therefore never want to see me again - joined soon after. An expert in consumer law as well, she instantly recognised the access to justice opportunity presented by this idea.
Soon after that joined Dr Ahmed Izzidien, a Senior AI Researcher at Cambridge.
Some close, talented friends and fellow Law graduates from Cambridge also joined me to provide further legal expertise to train the model: Elliot Wright, Ilsu Erdem Ari, Samuel Gerrard, Elliot Wright and Tiffany Chow. Today many of them are lawyers at leading City firms.
Dr Poorna Mysoor, also a Cambridge academic, was our IP expert who focused on legal and ethical compliance of our training data: she joined after being moved by our social mission.
Elizabeth Botsford, a veteran commercial lawyer from Baker McKenzie, joined to lead our legal data creation efforts.
This team began to form into a world-class legal and AI team. VCs started to notice us, with one commenting to me that the team was one of the most impressive they had ever seen in an early-stage startup.
I then raised more than £2 million in funding and expanded hiring even further.
The single most important hire was Dr Jonathan R. Schwarz, a Senior Research Scientist at DeepMind and Research Fellow at Harvard. He is a world leader in LLMs and brought lessons from Harvard Medical School, where he had been pioneering research into safe and robust models for medical use. He became Co-Founder and Chief Scientist.
Jonathan brought with him Professor Tom Hartvigsen from MIT and Fuad Issa from Cambridge, both of whom bolstered our status as a Tier 1 player in the Legal AI field and also joined our founding team.
We were also deeply fortunate to have a world-class business advisory team led by Terence Ng and Jonathan Barton, who brought decades of commercial experience from leading companies into the founding team.
I needed to assemble the world’s best team in the field, even though I had no business track record and no money.
I felt myself being funnelled into a conventional and highly predictable life path: one where my time and personal freedoms were at the mercy of the needs of the client.
Perhaps it’s because I’m an insufferable only child, but I care about my personal space and don’t particularly like being bossed around.
I also found some of the clients I dealt with very inspiring. They were taking risks, innovating, creating opportunities for others, and leading a life of their own choosing. That appealed to me.
There is also the financial aspect. I felt the need to create a path for myself that not only gave me independence but also would allow me, in my 20s, to retire myself, my parents and wider family. Few, if any, conventional routes offer this.
But above all else, it was the genuine wish to do something to democratise good legal advice.
Not everyone can afford A&O to solve their legal problem; and there are too many miscarriages of justice resulting from legal advice not being readily accessible to all people.
It was a challenge.
Throughout the life of the startup, I was responsible for growing and managing the team and providing monthly paycheques to an average of 14 employees (at one point rising to 35), for whom the startup was their full-time job.
I would have to manage disputes in my team, make complex commercial and strategic decisions on the spot, take on personal risk, shield my people from uncertainty and bad news, provide job security and financial support to all of them, and ensure that enough money continuously flowed into a very capital-intensive business where talent and compute are hugely expensive.
It was a lonely and sometimes overwhelming responsibility.
At the same time, I had to be a good trainee at A&O. This meant mentally boxing up and ignoring all the risk and uncertainty of the business, and being able to deliver on complex tasks that are expected of trainees at a Magic Circle firm like A&O.
My schedule looked something like this:
4.30am-9am: Wake up and work on startup
9am-1pm: Go to work at A&O
1-2pm: Eat lunch in the A&O canteen while having startup meetings on my phone
2pm-7pm: A&O work
7pm-10pm: Finish at A&O; work on startup
12am: Sleep and try not to have nightmares about going bankrupt or getting sued
Relationships, friendships and family suffered. But I knew it was the price I needed to pay if I was going to make this work.
A&O was very supportive and open-minded, and I’ll always be thankful to the firm.
The partners and senior lawyers I worked for took a pragmatic approach: so long as the startup did not interfere with my work for them, they would let me do it. There may have been a couple of brush-ins with some associates and HR, but thankfully there was always a partner to back me up.
Some of the partners found the whole startup story fascinating and enjoyed hearing about the highs and lows, living through the drama vicariously.
My fellow trainees found it very amusing to hear me having to hire, fire, and fire-fight during my lunch break in the canteen. Once the acquisition was announced, the trainee group chats did not disappoint.
Relationships, friendships and family suffered. But I knew it was the price I needed to pay if I was going to make this work.
Law Firm | Trainee First Year | Trainee Second Year | Newly Qualified (NQ) |
|---|---|---|---|
| A&O Shearman | £56,000 | £61,000 | £150,000 |
| Clifford Chance | £56,000 | £61,000 | £150,000 |
| Freshfields Bruckhaus Deringer | £56,000 | £61,000 | £150,000 |
| Linklaters | £56,000 | £61,000 | £150,000 |
| Slaughter and May | £56,000 | £61,000 | £150,000 |
Law Firm | Trainee First Year | Trainee Second Year | Newly Qualified (NQ) |
|---|---|---|---|
| A&O Shearman | £56,000 | £61,000 | £150,000 |
| Clifford Chance | £56,000 | £61,000 | £150,000 |
| Freshfields Bruckhaus Deringer | £56,000 | £61,000 | £150,000 |
| Linklaters | £56,000 | £61,000 | £150,000 |
| Slaughter and May | £56,000 | £61,000 | £150,000 |
Law Firm | Trainee First Year | Trainee Second Year | Newly Qualified (NQ) |
|---|---|---|---|
| Ashurst | £52,000 | £57,000 | £125,000 |
| Bryan Cave Leighton Paisner | £50,000 | £55,000 | £105,000 |
| Herbert Smith Freehills | £56,000 | £61,000 | £135,000 |
| Macfarlanes | £56,000 | £61,000 | £140,000 |
| Travers Smith | £54,000 | £59,000 | £120,000 |
When developing our AI model, we used leading law professors and commercial lawyers to adversarially test the model on areas of law on which these lawyers were the world leaders.
Yes, this sharpened our AI model. But something unexpected happened: the model also made fascinating connections and suggestions that the human brain, or received legal wisdom, would not typically make; and so the model became a healthy rival, a challenger, a devil’s advocate that caused our human lawyers to think in new ways, and ultimately become better lawyers.
I think that this is a blueprint of what a healthy relationship between lawyers and AI could look like.
But first, let’s look at the current capabilities of AI in the legal realm.
Experiments starting in the last decade have shown how AI systems can outperform and radically outpace lawyers on a range of tasks.
The seminal study was conducted by Israeli startup LawGeex in 2018 where an AI model identified 20% more problematic clauses in a sample set of NDAs than human lawyers did–and the AI model did this 3,000 times faster.
AI models have since advanced in power and capabilities at a near exponential rate, though from Big Tech providers we are seeing diminishing increases in model performance in some areas. The general trend, however, is that virtually all areas of human activity will be more or less replicable by AI models in the medium-term.
It is therefore not a question of whether AI models will transform the legal industry, but instead how we choose to channel that transformation.
We can think of AI as a moving train: it’s arriving at the station. We as lawyers cannot control whether that train leaves the station or where that train goes. All that we can control is how we react to it: whether we choose to get on the train, or be left at the station.
We need to view AI models as force multipliers, efficiency drivers, helpful companions and healthy competition. They are not here to replace lawyers but to enhance them. AI won’t fully replace lawyers; but lawyers who use AI will replace those who don’t.
A majority of trainee and lower-level associate tasks can already be performed to a high standard of accuracy by existing AI models. It is arguable that the ‘instinct’ experienced associates and partners develop over the course of their career is less exposed to disruption by AI, but it is certainly not invulnerable.
AI won’t fully replace lawyers; but lawyers who use AI will replace those who don’t.
The five core obstacles to AI disrupting the legal industry are, in my view:
There are solutions to all of these, and I spend a lot of time speaking with partners (mostly from US firms) about their thinking on these questions.
Make no mistake: there are groups of partners from all the major law firms meeting and discussing these problems behind closed doors, sometimes in meetings I’m privy to. They know which way the wind is blowing and they know that the commercial survival of their firms relies on proactiveness and agility in the face of this transformative technology.
I would suggest that the next 5-10 years will therefore see major law firms that pivot to being predominantly AI-powered, with a great range of tasks fully or almost fully automated.
We will see new disruptor law firms that emerge that are ‘AI first, human second’, and they will use the scale offered by AI to try to capture previously inaccessible areas of the legal market. There will still be humans in the loop for most but not all tasks, and there will be novel ways of billing clients and apportioning risk.
We will only arrive at a stage where an AI model can be fully trusted (or trusted as much as a human is) once the empirical evidence supports that proposition and once that is understood and accepted in mainstream industry thinking, and we are not there yet on many legal tasks. This will happen in the next 2-3 years.
We will see major firms commissioning custom LLMs trained on their in-house precedents, client profiles and other data, and achieving efficiency savings that will lead to a reduction in the need for manpower and a sharper focus on hiring the ‘very best’ candidates.
We will see a rivalry between Foundational Legal LLMs (e.g. the Safe Sign model, which is built from the ground up to serve legal needs) and general purpose LLMs (e.g. the ChatGPTs of the world): I am firmly behind smaller, purpose built models like Safe Sign’s continuing to outmanoeuvre and outpace big, clunky, generalist models on legal tasks – but only time will tell whether this hypothesis is correct.
Lawyers will also need to develop ‘bullsh*t sensors’: firms are already flooded with products that are nicely marketed but technologically unimpressive and ultimately disappointing; the legal industry will need to develop an instinct and basic technical knowledge base to be able to discern the utility of AI systems and avoid being taken for a ride by the sea of startups we are seeing emerge overnight.
At a deeper level, we will see a realignment of the relationship between lawyers and AI: mistrust turning to cautious comfort, turning to deference at the risk of complacency.
We will need to develop ways of ensuring human lawyers retain sufficient knowledge and attention to detail to continue to actively practice law and verify AI model outputs - and marking homework written by a super-intelligent system can sometimes be more difficult than doing the homework yourself.
What is most exciting for me is that, as we get closer to a genuinely trustworthy ‘AI Lawyer’, we also get closer to a system that will give everyone on the planet the ability to understand the law and enforce their rights under the law in a way that has never been done before. We will well and truly achieve ‘equality before the law’.
We will see new disruptor law firms that emerge that are ‘AI first, human second’.
Law Firm | Trainee First Year | Trainee Second Year | Newly Qualified (NQ) |
|---|---|---|---|
| Ashurst | £52,000 | £57,000 | £125,000 |
| Bryan Cave Leighton Paisner | £50,000 | £55,000 | £105,000 |
| Herbert Smith Freehills | £56,000 | £61,000 | £135,000 |
| Macfarlanes | £56,000 | £61,000 | £140,000 |
| Travers Smith | £54,000 | £59,000 | £120,000 |
There came a point in late 2023 when our scientific team had achieved momentum.
We had a critical mass of funding and world-leading expertise that allowed us to get things done much more quickly than our rivals, and to a higher standard. We had built the best team in the world for what we were trying to achieve.
Then, in early 2024, we conducted rigorous internal evaluations that indicated that our model was the best in the world on a wide range of complex legal tasks.
We released an industry announcement of our results and prepared for a $15 million Series A round. VCs swooped. So did Thomson Reuters - their key venture fund manager replying within minutes to our newsletter.
Thomson Reuters are the world’s leading provider of legal technology and information (responsible for Westlaw and Practical Law, the world’s most popular legal products).
They initially proposed leading the round. Once we got into due diligence, however, they realised that our scientific objectives were pivotal to their strategy and that our technology was truly ahead of all others they had looked at.
They proposed an acquisition. We refused initially.
Eventually, though, the offer became attractive enough to take into account the opportunity cost of exiting so early. Our scientists and lawyers were offered leading positions at Thomson Reuters and the opportunity to pursue their research with a budget that is larger than the defence budget of some countries.
Within three months of the first conversation, we had sold the company. I was still a trainee at A&O at the time, and had never worked on the startup full-time.
It was possible only due to the efforts of my team, in particular Jonathan Schwarz, who demonstrated beyond any doubt that he and the team he led in Safe Sign were the world’s best.
The deal was very much in the ‘beyond’ range.
It has changed the lives of everyone involved. It is well and truly a ‘dream come true’. For some context, the deal came out of Thomson Reuters' $8 billion war chest for winning the AI race.
We’re the only pre-revenue business Thomson Reuters has bought in its 174-year history, so this was a big milestone for them, as well as for us.
For me personally, having gone through all the ups and downs of the business, shouldering its legal and financial risks, while balancing a training contract at A&O, I could not be happier with the outcome.
The deal changed the lives of everyone involved.
For my sanity, yes.
My team and I have joined Thomson Reuters pursuant to an earn-out under the Share Purchase Agreement.
Jonathan Schwarz is Head of AI Research at the company; I am Senior Director, Thomson Reuters Labs; and our other key team members occupy senior positions in the technological arm of the company.
We continue to make strides in maintaining our lead as the world’s best Legal AI model and are integrating our technology to power Westlaw and Practical Law - products that are used by almost a million lawyers worldwide and 97% of all law firms.
Those products contain 174 years’ worth of high-quality, painstakingly created legal content, and likely constitute the most irreplicable and comprehensive legal dataset on the planet: the core ingredient of an industry-dominant AI model.
I’ve always believed in finishing something once I’ve started. So I won’t rest until we have provided lawyers around the world with the first AI model they can truly rely on, thereby triggering the transformation of the legal industry into the AI age.
In the much longer term, I don’t rule out continuing on the entrepreneurial route.
It is fundamentally satisfying to have an idea, build a talented team around it, win over investors, face odds uncounted, laugh in the face of immense difficulties, and, with some luck and a lot of sweat, deliver a successful outcome for those who have put their faith in you.
Get the free email that keeps UK lawyers ahead on the stories that matter.
We send a short summary of the biggest legal industry and business stories you need to know about three times a week. Free to join. Unsubscribe at any time.
Copied to clipboard!
