The Magic Circle trainee who built an AI startup in his spare time — and then sold it to a tech giant

Alexander Kardos-Nyheim on juggling a training contract with launching a business, and what the future holds for lawyers in an AI-driven world

Alexander Kardos-Nyheim made headlines when he sold his legal tech startup, Safe Sign Technologies, to Thomson Reuters in 2024. The most intriguing part? He was still a trainee at A&O Shearman at the time. Speaking ahead of his appearance at LegalEdCon 2026 next month, I started with the obvious question: what possessed him to build and launch an AI startup while still in the middle of a training contract?

Alexander deftly sidesteps the question, launching instead into the startup’s origins. “The reason I started the startup was the same reason why I decided to go into law in the first place.” As a teenager, he says, he “had to defend my community against demolition.” With limited means and no easy access to lawyers, he found himself learning the law “alongside my GCSEs”, teaching himself enough to help his mother and neighbours fight back against property developers.

It culminated in a town hall showdown. “It was me on one side and the Magic Circle lawyers of the developers on the other.” After Alexander gave his pitch, those around him feared the worst. Instead, applause broke out across the room, and the councillors eventually ruled in their favour.

That experience made the law feel both powerful and unevenly distributed. “With the law, if you work at it, if you’re motivated by what you’re doing, and if you work hard, you can achieve very important things for people,” he tells. Years later, that memory evolved into a different sort of question. “What if technology could help more people get to that point?”

For Alexander, the answer was AI. The mission became, in his words, “to democratise the law using AI”. But, he adds, that comes with a “big responsibility”, because he “wanted something reliable enough for people to defend themselves in court or to protect their rights”.

That’s where Alexander draws a line between legal AI and general-purpose chatbots. “ChatGPT and the big models are not trained for these very high stakes, accuracy, critical use cases.” In his view, big tech tends to prioritise coverage and capability across lots of tasks, meaning “their focus on reliability, safety, robustness is actually lower down on their priority list”.

So he set out to build a legal large language model (LLM) engineered for accuracy, which manifested eventually as Safe Sign. Alexander is candid about his own background:

“I had no experience in tech whatsoever,” he tells me. “I have never written a line of code. But I was lucky to be joined by a team of world-leading AI scientists and engineers from Cambridge, MIT, Harvard and Google DeepMind.”

The result, he claims, was “an LLM specially trained on law” that was “the most reliable in the world for a broad range of legal questions.” Their expertise, he added, led to the technology being “recognised as best-in-class”. His summary of the company’s position is blunt. “We are the serious response to ChatGPT, for law.”

Then came the hard part, turning it into a sustainable business. This meant juggling an already demanding Magic Circle training contract, and the hours that come with it, alongside launching his very own company. Alexander was blunt about it. “There are 24 hours in a day. I used them.”

Of his time at A&O Shearman, Alexander is genuinely grateful, particularly for the skills and experience it equipped him with, as well as the exposure it gave him to the world of legal tech. “The firm gave me deep insight into the cutting-edge of legal tech,” he says. “I sat as a trainee in partner David Wakeling’s Markets Innovation Group, the first group of lawyers in the world to adopt AI tool Harvey.” He also cites a partnership and supervisors which proved genuinely supportive, making special reference to his supervisor Sunil Mawkin, now partner at CMS, who “provided unparalleled understanding and support in my final seat.”

He is candid, though, about the aspects of trainee work that ultimately did not appeal to him, useful as they may have been. Much of it, he felt, was “super detail oriented, but quite repetitive and the sorts of things that might be automated soon by AI.”
For Alexander, rote tasks carry an inherent ceiling. Do them perfectly and you have done exactly what was required; get one detail wrong and you have “let the side down.” An AI tool, by contrast, has no such fixed upper limit. “It’s not often there is a ceiling as to how good a product can be,” he says, before acknowledging the flip side. “There’s no floor as to how bad it can be.”

When it came to launching the business itself, it was no easy task. Whilst Alexander may have received some “informal advice” from A&O partners when it came to Safe Sign’s eventual acquisition, the road there was far from smooth and ultimately an independent endeavour undertaken by Alexander and his team.

Alexander says they initially tried, and failed, to launch the product direct to consumers. “We did not have the investors who wanted to support that,” he explains. To keep the business alive, he ultimately resigned himself to shelving that approach “at least temporarily”, while continuing his training contract.

Over the course of this period, Alexander says the business came close to bankruptcy repeatedly, close enough, as he puts it, for “the light bulbs to start to flicker”.
Eventually, however, the startup’s research started to attract attention within the market. Alexander says they published papers at “some of the top AI conferences”, and that Thomson Reuters got in touch during a Series A raise. That potential investment culminated in the acquisition that made headlines across legal publications and beyond.

Today, Alexander co-leads AI research at Thomson Reuters globally, operating as an independent entity inside the company and reporting directly to the board and the president. His mandate is to create “the world’s most reliable large language model for law”, with a clear focus on building a system that “doesn’t make stuff up”, even if it “might not be the most imaginative all the time”.

He outlines the scale too. An AI research lab within Thomson Reuters with around 150 lawyers and 40 to 50 scientists, competing directly with the likes of Anthropic, Google DeepMind and OpenAI. And it seems to be paying off, with the team gearing up to release their Safe Sign LLM, now rebranded as Thomson, to market.

So what does all this mean for trainees and junior lawyers watching AI move from curiosity to workplace expectation?

Alexander doesn’t think there will be one single effect. Some firms will treat AI mainly as a cost cutter. Others will see it as an “enabler”, a way to shift juniors towards more interesting work sooner, alongside automating the repetitive tasks.

He predicts new junior roles too: “hybrids between lawyers and technologists”. But expectations are changing fast. Alexander says partners now expect trainees and associates to deliver the first draft of work “in hours rather than days”, and for it to be “better than it was in the pre-LLM phase”.

There are trade-offs. When I ask about the worry that juniors might lose something if more repetitive tasks disappear, Alexander doesn’t laugh it off. “It is a risk.” Certain cognitive skills will, he warns, “fall away”.

To wrap up, I ask what juniors and aspiring lawyers should focus their attention on now. Plenty of people, Alexander says, will tell you which AI tool is best at what, but he sees too much “superficial stuff” and too many “soundbites”. His advice is to go deeper.

“Junior lawyers should grapple with the fundamentals, understanding what a large language model actually is, how pre-training, fine-tuning and reinforcement learning work, and how these models are improved over time,” he says. |Bring that understanding into practice and you’re automatically making yourself extremely valuable”.

Alexander Kardos-Nyheim will be delivering the keynote at LegalEdCon 2026 at Kings Place, London on Thursday 14 May 2026.

LegalEdCon is headline sponsored by BPP University Law School, BARBRI and The University of Law, with The College of Legal Practice as a silver sponsor.

Final Release tickets are now available. Secure your place.

This event is targeted at legal academics and those involved in learning & development and graduate recruitment at law firms. It is not a student event. However, we do have a limited amount of free tickets for our student campus ambassadors. You can find out more about the campus ambassador programme here.