Solving Security Problems With AI And Experience
“It’s fascinating but difficult to make an AI model that can solve complex, real-world problems, and even harder to help it make useful recommendations and decisions that lead to a good outcome,” said Zhenwen Dai, Co-Founder and CTO of Trent AI. Yet, Trent AI’s leadership team is perfectly placed to do just that, because its members all have a strong track record in solving AI-related problems and applying them to deliver solutions that delight customers.
Zhenwen, Neil Lawrence, and Eno Thereska founded Trent AI to help make every software product secure by design and tackle a new security challenge. AI-generated software is increasingly used by businesses to build the apps, sites, and computer infrastructure we depend on for work and play, but it is likely to rapidly deploying human errors at large scale presenting security risks.
Trent AI’s products will ensure that organizations adopting agentic workflows and autonomous systems can embed safety and compliance from the start, delivering invisible automatic security for the AI age.
Meet The Leadership Team
Neil Lawrence, Chief Scientist of Trent AI, began his career working on oil rigs as a wireline logging engineer having completed an undergraduate degree in Mechanical Engineering. “I’d hoped to be sent on adventurous jobs in Colombia and Nigeria, but I ended up in Morecambe and Great Yarmouth,” he joked.
An article in the New Scientist about neural networks captured his imagination. “I thought that certainly sounded cool,” he said. Neil saw complex problems on the rigs that couldn’t be solved with existing technologies. “I began coding neural networks when I wasn’t needed on the drill floor,” he explained. While Neil developed a great respect for his colleagues, he knew he wanted to work on neural networks. “Giving up the large salary to do my PhD at Cambridge was one of the best decisions I ever made,” he said.
Neil now has more than 25 years of experience with machine learning models and the real-world applications of this technology, and prior to co-founding Trent AI, was director of machine learning at Amazon where he learnt “an enormous amount about leadership, decision-making and running a business”.
Having been the first PhD student to work on machine learning at the Cambridge Computer Lab, Neil is now the inaugural DeepMind Professor of Machine Learning at the University of Cambridge, where he continues to work on research at the crux of machine learning and systems. He is also a visiting professor at the University of Sheffield and serves on the board of the conference AISTATS and the ELLIS Foundation. These positions, as well as being series editor for the journal Proceedings of Machine Learning Research, give him a unique view into potential AI advances and security problems, enabling him to help Trent AI develop the best solutions.
Neil suggested “Trent” as the name for the company having drawn inspiration from the River Trent, which he described as “a network of the Industrial Revolution” because it enabled the efficient and high-capacity movement of raw materials and finished goods to and from the factories. Trent AI reflects the founders’ intention to build infrastructure that enables progress and amplifies the efforts of businesses, with the potential to transform industries.
Zhenwen Dai, CTO at Trent AI, taught himself computer programming at high school and “somehow got fascinated with the idea of artificial intelligence despite it being a period of AI Winter when nobody talked about artificial intelligence.” There were few people working in the field when he started university and “nobody knew whether AI could become popular”. The most related subject he could study at the time was data mining and then computer vision, but he was determined to pursue his “personal interest” and finally got the chance during his PhD. Zhenwen then conducted postdoctoral research with Neil and they co-founded Inferentia, which was acquired by Amazon where he worked as a machine learning scientist before joining Spotify.
Spotify aims to help users find the music they love and discover new songs, but of course no two listeners are the same, and generating useful recommendations that evolve with their unique taste is challenging. Zhenwen led a research lab at Spotify, developing AI for personalized listening experiences. “It’s a particularly interesting domain for machine learning people, because there are few consequences for not giving a good song recommendation. Users are usually tolerant of the decision, which allows us to develop very different types of machine learning models, and try to see how we can optimize user experience with machine learning models,” he explained.
Zhenwen worked with product teams to test some of the algorithms, which “led to many interesting real-world learnings.” He is using a similar approach at Trent AI and has more than 15 years of experience in the machine learning space.
Eno Thereska, CEO at Trent AI, worked on one of the first autonomous data centers while studying for his PhD at Carnegie Mellon University. Having taken the famous Machine Learning graduate level class with Tom Mitchell and Andrew Moore, he began applying AI straightaway (and with a certain degree of naivety required) to systems. “AI is just a tool. It helps you solve problems, but there’s a whole lot more that needs to go into making systems autonomous,” he said.
He now has more than 20 years of experience in building secure, scalable, and autonomous systems, having been an early engineer at Confluent, Principal Engineer at AWS and prior to co-founding Trent AI, a Distinguished Engineer at Alcion, which was acquired by Veeam. “Twenty years ago, I was using decision trees and queuing theory, and today I’m using LLMs and reinforcement training, but some of the basic problems are the same because AI is both powerful but also makes subtle mistakes,” he said.
Eno worked with Christoph on AWS’ S3 Intelligent-Tiering; a storage class that automatically optimizes costs by moving data between frequent and infrequent access tiers based on access patterns. “The idea was to use AI to predict how files are used and automatically move them somewhere cheaper to delight the customer with a lower bill.” The distributed storage system is now one of the largest in the world and a multibillion-dollar business.
Christoph Bartenstein is Chief Product Officer at Trent AI. He developed an interest in AI while in business school and working in tech companies, helping customers succeed by solving specific problems, such as storage costs while working at AWS. “For me, it wasn’t about creating a fancy AI product. It was about the end result: saving customers money or getting better performance for them,” he explained. Having talked to customers, he understood they didn’t care too much about storage per se, but wanted performance at a good price point so they could focus on their businesses.
Christoph has over 20 years of experience leading product and engineering teams at high-growth, global technology companies in the US and Europe. He held various leadership roles at Microsoft as well as AWS, worked in Technology Consulting and co-founded two startups, which gave him experience in building from zero and putting products in customers’ hands.
“I thought about AI as one tool among others used to drive incremental change, but it’s more than that; a new paradigm that changes how business works.” Now leading product development at Trent AI, he believes there is an opportunity to help businesses that are facing new security challenges brought about by the adoption of AI.
Paz Klapztein, Head of Operations at Trent AI, has around 10 years of experience in Operations, and joined the company from Spotify, where she worked closely with Zhenwen. When he asked her to join Trent AI at an early stage, she didn’t hesitate because she knew that his technical expertise would lead to a high-quality product. “It’s the right time for this kind of company,” she said. “Not just because of where AI is, but because it’s a chance to build the operational foundations right from the start.”
At Spotify, Paz helped enable Zhenwen’s team by improving planning, execution, and cross-collaboration. “That role gave me a deep understanding of the constant change teams face, and the clarity and flexibility needed in operations to scale organizations not only fast, but also well, and with the right culture,” she explained. She has now brought her skills to Trent AI, with a focus on shaping how the organization functions as it scales, building the structures and systems that allow Trent AI to operate with clarity, resilience, and momentum. “I’m comfortable with complexity, and I thrive on bringing clarity,” she said, and is excited to embrace the operational challenges of scaling Trent AI.
The Biggest AI Challenges
“We already know that organizations big and small can struggle to protect their data, with cyber security experts only accessible to larger companies,” said Neil. “AI technologies cannot and should not replace experts, but we would like to support them in identifying fast-moving threats and prioritizing the threats to tackle first.”
One area where experts may need support is in dealing with new problems around prompt injection; a cybersecurity exploit in which bad actors can craft inputs that appear legitimate but are designed to cause unintended behavior in large language models.
“AI generated code or vibe prompting (which enables the creation of software from natural language instructions) is another area which has changed the nature of getting nasty code into software,” Neil said.
While AI generated code arguably democratizes the writing of software, the person using this new technique may lack the expertise to scrutinize the code, debug, or maintain it, leading to an increased risk of accidentally introducing security vulnerabilities.
“Machine learning is a fundamental break of the data code barrier, which is the most fundamental barrier in security”, Neil explained. “Understanding how to help with potential problems that’s going to cause is important, and this difficult problem is something that Trent is working on solving.”
Eno believes that passing on large context to an AI agent is a key test. “Sometimes it can be expertise accumulated over years, so how you transmit that quickly and practically in the security domain is difficult,” he said.
Zhenwen considers how to train specialist AI agents to solve tasks reliably to be one of the most interesting challenges. “In the longer term what’s really interesting to me is how we can build an AI system that can improve itself and learn efficiently from its past experiences, removing a lot of manual effort in training.”
One difficulty to overcome is how to make AI helpful rather than overwhelming, according to Paz. “It’s in everything we do and use now, so knowing how to safely integrate it into workflows and use it responsibly is key,” she said.
Christoph believes pinpointing how companies can create value from AI is an important area of focus. “Once you figure it out, it’s astonishing because it’s not about driving incremental change. It’s a complete shift.” He believes that companies using Trent AI will experience this shift. “It’s not about doing a static code scan better than what’s already out there. It’s changing the way security works for software companies,” he explained.
Customer Feedback and Disruptive Potential
Early feedback from potential customers has been great and there has been lots of testing and internal discussions about how to develop the most helpful product possible. “I’m excited everyday to talk to customers and see what they like, and what they don’t like,” said Christoph.
“Inevitably, there will always be some aspect you haven’t thought about”, Zhenwen said. “As technology developers we’re keen to be exposed to real world use cases for our AI products, especially those that are harder to imagine pre-launch, and when we build the next version, these use cases will guide our innovation because we will be able to see where the technology can really make a difference,” he added.
Zhenwen is sure that building specialist AI agents will deliver value to Trent AI’s customers, while Eno is certain the cyber security field is “ripe for disruption”. Trent AI’s aim is to help companies design and build secure code, while identifying ongoing vulnerabilities. “The Holy Grail is code being automatically updated and staying continuously secure over time,” he added.
Trent AI is product focused, and serving customers’ needs is at the heart of the business. Paz is excited to support customers, as well as putting the right internal systems in place that will enable Trent AI to scale quickly and sustainably, so it can bring its products to more businesses and build new ones.
Research will inform the company’s direction in a market that is rapidly changing. Christoph explained that the combination of research, which has a long-term horizon and highlights what might be possible, combined with customer feedback, which is great for shorter and more iterative innovation, gives you the perfect product perspective.
“No one knows exactly where AI will take businesses next, but their software will certainly need to be secure,” said Christoph. “There are not many times in history where you have the opportunity to shape entire markets and business models. This is one of those opportunities.”
