Most startup founders chase revenue charts, headline funding rounds and polished stage presentations.
asserted
founders → chase → presentations
Alexander Kardos-Nyheim took a different route.
asserted
Nyheim → take → route
While training as a solicitor at Allen & Overy, he built Safe Sign Technologies, a legal-specific large language model (LLM) startup designed to make artificial intelligence more reliable for legal work.
asserted
intelligence → train → work
Roughly two and a half years after its founding, Thomson Reuters acquired the company before it had generated revenue from subscriptions.
asserted
it → acquire → subscriptions
The acquisition, announced on 21 August 2024, demonstrated how specialised research and technical expertise can create value before a business has paying customers.
asserted
business → announce → customers
The story raises a question every founder, investor and legal professional should consider, what truly makes a young company valuable?
asserted
company → raise → question
From a teenage planning dispute to an unusual acquisition, Kardos-Nyheim’s journey offers a closer look at the changing economics of artificial intelligence.
asserted
journey → offer → intelligence
How a teenage planning dispute at Rollins House in south-east London shaped a future startup founder
asserted
dispute → shape → founder
The story begins long before any startup existed.
asserted
startup → begin → ?
According to a first-person commentary published by Fortune, the house where Kardos-Nyheim grew up, Rollins House, was a run-down Art Deco former factory in south-east London.
uncertain
Nyheim → accord → London
His mother lived there, and a developer wanted to redevelop the site into flats.
asserted
developer → live → flats
The family could not afford lawyers, so, while still at school, he taught himself planning law between bouts of exam revision.
uncertain
he → afford → revision
He examined records that others had overlooked and assembled a dossier of precedents.
asserted
others → examine → precedents
Facing a council panel across the table from an experienced law firm, he presented the family’s case, and they won.
asserted
they → face → case
The experience taught him the value of detailed research, persistence and challenging assumptions, lessons that would later influence his approach to legal technology.
asserted
that → teach → technology
Training as a solicitor at Allen & Overy by day and building a startup at dawn and every night
Years later, that determination was tested in a far more demanding setting.
asserted
determination → train → setting
Kardos-Nyheim trained as a solicitor at Allen & Overy while developing Safe Sign Technologies outside working hours.
asserted
Nyheim → train → hours
He began at half past four in the morning, working on the company before a full day of legal training, then returned to the startup each evening.
asserted
He → begin → startup
Separately, the company often came close to running out of cash and needed new funding to keep going.
asserted
company → come → funding
Its initial idea, a consumer-facing legal product, also failed to attract investors, forcing the team to rethink its direction.
asserted
idea → face → direction
Instead of building another legal service, the company turned towards the underlying technology, artificial intelligence models trained specifically for legal reasoning and research.
asserted
company → build → reasoning
Kardos-Nyheim’s background combined legal training with an understanding of the difficulties faced by people trying to navigate complicated rules and documents.
asserted
background → combine → rules
That combination became central to Safe Sign’s identity.
asserted
combination → become → identity
The Law Gazette’s report on the acquisition identifies him as a Cambridge law graduate and former trainee solicitor at A&O Shearman, the firm formed through the combination of Allen & Overy and Shearman & Sterling.
asserted
report → identify → Allen
Alexander Kardos-Nyheim (Image: City of London School)
Securing capital from North American investors after British investors repeatedly said no
Funding proved equally difficult.
asserted
Funding → secure → investors
Local investors repeatedly turned him down, according to his account in Fortune, so Kardos-Nyheim travelled to New York with only £200 to his name.
uncertain
Nyheim → turn → name
The effort paid off, with nearly all of the company’s capital eventually coming from North America rather than the UK.
asserted
all → pay → UK
He observed a difference in the questions investors asked, American backers were more likely to explore how they could help, while British investors tended to focus on how the venture might fail.
uncertain
venture → observe → questions
Once the company had enough funding to continue, the team abandoned the immediate pursuit of revenue and concentrated on building its proprietary artificial intelligence models.
asserted
team → have → models
Its researchers were drawn from institutions including the University of Cambridge, the Massachusetts Institute of Technology, Harvard University and DeepMind.
asserted
researchers → draw → Technology
The objective was not simply to add an AI chatbot to an existing legal service, but to develop models specifically for the demands of legal work.
asserted
objective → add → work
This was a difficult proposition to sell.
asserted
This → sell → ?
Developing specialised LLMs requires considerable research, testing and technical expertise, while commercial returns can take time to emerge.
asserted
returns → develop → time
Investors expecting rapid sales were therefore less receptive to a business whose principal asset was still under development; by Kardos-Nyheim’s account, most lost interest.
asserted
most → expect → interest
Safe Sign nevertheless continued to prioritise the technology, betting that reliability and domain-specific performance would eventually distinguish it from general-purpose AI systems.
asserted
reliability → continue → systems
Thomson Reuters acquires a pre-revenue artificial intelligence startup in an unusual deal
The research-led strategy ultimately attracted a major buyer.
asserted
strategy → acquire → buyer
On 21 August 2024, Thomson Reuters announced its acquisition of Safe Sign Technologies, describing the UK business as a developer of legal-specific large language models.
asserted
Reuters → announce → models
Based on its own internal assessment, the company said Safe Sign’s models had demonstrated industry-leading performance in domain-specific evaluations, and that the acquisition would help accelerate the development of its CoCounsel AI assistant.
asserted
acquisition → base → assistant
The price was not disclosed.
asserted
price → disclose → ?
The deal was particularly notable because Safe Sign had no subscription revenue when it was acquired.
asserted
it → have → revenue
…and 21 more, not listed.
The Aporia
