A wrong answer from an AI chatbot is one thing. An AI agent leaking internal information or bypassing security controls is another. That distinction led Joan Vendrell, co-founder and CEO of NeuralTrust, to build a platform that helps enterprises monitor, secure, and govern AI agents connected to their business systems. Drawing on his experience at Mango, Amazon, and McKinsey, Vendrell saw a gap between the speed of AI adoption and the controls companies needed to manage its risks.
In this interview with Eqvista, Vendrell shares the journey from that early insight to a $20 million seed round, including the challenges of winning enterprise customers and building from Barcelona. He also discusses NeuralTrust’s approach to AI security, the equity decisions that helped establish its foundation, and what he would do differently when planning for international growth.
What was the moment that led you to found NeuralTrust? What problem in AI security were you seeing that others weren’t talking about yet?
Before NeuralTrust, I was Chief Data Officer at Mango, and before that I worked at Amazon and McKinsey. What I kept seeing was a gap nobody was addressing head on: companies were racing to connect AI agents to their real systems, email, internal databases, customer tools, without any layer that actually inspected what those agents were doing in real time. A wrong answer from a chatbot is annoying. An agent connected to your email system leaking internal information, or a tool-calling workflow that gets manipulated into bypassing controls, is a different category of risk entirely. Traditional security tools like firewalls and SIEMs weren’t built to understand multiturn conversations or tool misuse. That gap, between how fast agents were being deployed and how little visibility or control enterprises had over them, is what led me to start NeuralTrust with Victor and Alejandro.
Every founder has a story of early doubt or near-miss. Can you share a moment in those first 6-12 months when you questioned whether NeuralTrust would make it, and what kept you going?
The hardest stretch wasn’t a single dramatic near-miss, it was the quiet grind of the first several months, when we had a strong technical thesis and almost nothing else. No signed customers yet, no analyst validation, just a small team convinced that enterprises were about to have a problem most people hadn’t named yet. There were weeks where I’d leave a meeting having pitched the vision well and still walked away with nothing signed, and I’d genuinely wonder if we were early in the good sense or early in the sense that matters more, too early for anyone to actually pay for it. What kept me going was narrow but concrete: the first few conversations with security and risk teams at large enterprises where I didn’t have to convince them the problem existed. They already knew. They just didn’t have a vendor yet. That was the signal that told me we weren’t chasing a hypothetical, we were early to something real, and being early to something real is a very different bet than being early to something imagined.
Barcelona isn’t traditionally known as a cybersecurity hub. How has building in Spain shaped your approach to hiring, fundraising, and product development?
Building from Barcelona has actually worked in our favor more than people expect. On fundraising, we brought in European investors like Alstin Capital, VentureFriends, Seaya, and Kibo Ventures, alongside institutional and public backing from the European Innovation Council and Spain’s State Research Agency, support that isn’t as readily available to a US-based startup in the same way. On product and positioning, being headquartered in Europe matters directly to our customers. With the EU AI Act tightening compliance requirements, a meaningful share of our government and banking customers are actively looking for vendors headquartered outside the US for reasons of technological and defense sovereignty. Being Barcelona and London based, and now with a presence in Munich and New York, lets us tell a credible European-first story to exactly the customers who need to hear it. On hiring, Barcelona has a strong, underused pool of technical talent that hasn’t been picked over the way Bay Area or London talent has.
Congratulations on closing the seed round at $20M. Could you share your fundraising journey, how did you position NeuralTrust to investors, and what milestones gave them confidence at such an early stage?
We closed $20M, the largest cybersecurity seed round raised by an EU company to date, led by Alstin Capital, with VentureFriends, Seaya, Kibo Ventures, Banc Sabadell, EA Ventures Plug and Play Fund, and Finaves (IESE Business School’s venture fund) participating, plus public backing from the European Innovation Council and Spain’s AEI. We positioned NeuralTrust not as a point solution but as the infrastructure layer that makes AI agent adoption measurable, governable, and safe, the same category-defining role that endpoint security and cloud security played in their respective eras. What gave investors confidence wasn’t a pitch deck alone. It was that 92% of our customers already generate over $1B in annual revenue, real logos like Iberia, Air Europa, Abanca, and Banc Sabadell, our platform inspecting millions of agent interactions daily with roughly 1.2% flagged for genuine intervention, and independent validation from Gartner (Representative Vendor in two Market Guides), KuppingerCole (Leader in the 2025 Generative AI Defense Leadership Compass), and MarketsandMarkets (Leader in the 2026 Agentic AI Security Quadrant). We also contributed original research to the field, identifying threat techniques like Echo Chamber and Semantic Chaining that are now part of the OWASP AI Security Project taxonomy, which signaled we weren’t just building product, we were shaping the category.
As you scaled from idea to seed, how did you manage the increasing complexity of your cap table and equity distribution among co-founders, advisors, and early employees?
We kept the founder structure simple and clean from day one, so that as we added advisors and early hires, we were building on top of something everyone already understood rather than renegotiating it every round. Victor, Alejandro, and I split founder equity early and deliberately, before there was any funding pressure to complicate the conversation, and we set standard vesting from the start so there was never any ambiguity about commitment. As we brought on advisors, we kept those grants small and milestone-based rather than open-ended, and we built a proper option pool ahead of the seed round rather than carving it out under pressure during the raise, which is where a lot of founders lose leverage. The complexity really showed up once we started hiring internationally, across Spain, the UK, Germany, and the US, since equity mechanics and tax treatment differ meaningfully across those jurisdictions. That’s where having proper legal and equity management support early, rather than treating it as an afterthought, saved us from having to unwind bad decisions later.
Looking back, is there anything you’d do differently in your incorporation or early equity planning if you could start over?
Yes. I’d set up the international structure earlier rather than treating it as something to solve once we actually had a reason to hire in the UK, Germany, or the US. We got the founder equity and vesting right early, which I’m glad we did, but the multi-jurisdiction complexity, different equity vehicles, different tax treatment, different compliance obligations across Spain, the UK, Germany, and the US, is something I’d have mapped out with proper counsel from the start rather than building incrementally as each new office came online. It wasn’t a costly mistake, but it added avoidable overhead at moments when we had more urgent things to focus on, like closing customers and closing the round.
NeuralTrust offers four core products: TrustGate, TrustGuard, TrustLens, and TrustTest. How did you decide on this product architecture, and what’s been the biggest surprise in how customers actually use them?
The architecture followed the actual lifecycle of an AI agent in production. TrustGate sits at the gateway layer, managing and controlling every interaction between models, external tools, and enterprise systems, inspecting requests before they execute. TrustGuard provides runtime security, detecting and stopping threats as agents operate. TrustLens gives visibility into posture and behavior over time, so security and compliance teams can see what agents are actually doing across the environment, not just what they’re configured to do. TrustTest rounds it out with red teaming, so customers can find the gaps before an attacker does rather than after. The biggest surprise has been how much customers lean on TrustLens for internal accountability, not just external threat defense. A lot of enterprise buyers came to us worried about attackers.What they actually spend most of their time using our platform for is proving to their own risk and compliance teams that their agents are behaving as intended.
The AI security landscape is getting crowded fast. What’s your view on the competitive moat for NeuralTrust, and how do you think about defensibility beyond just being first to Market?
Being first doesn’t mean much on its own in a market this fast-moving, so we’ve built our defensibility into the product itself rather than relying on being early. A few things make a real difference here. We’re the only gateway with security models built directly into the product, direct and indirect prompt injection detection, toxicity, file inspection, and more, so customers aren’t forced to bolt on a separate guardrails solution the way they are with alternatives that only offer basic regex matching and third-party integrations. We also ship with a catalog of over 200 MCPs already built in. Most gateways, including well-known names, give you compatibility with MCPs in theory but leave you to set each one up and maintain it manually, which becomes a real burden at scale. Our gateway also has stateful memory built in, so unlike gateways that treat every request as independent, we remember what a user has done before and use that history to make better security and routing decisions at the moment. On top of that, we’re one of the few gateways that propagate identity all the way through the stack, so a customer’s IdP authentication doesn’t just get you into the gateway, it also determines model and MCP routing and authenticates directly into the MCP servers themselves. And all of this runs at genuinely leading performance, 20,000 requests per second per node with linear scalability, alongside security that measurably outperforms alternatives, F1 scores above 90%, roughly 20 to 30% better than what else is out there. Put together, that’s a moat built from architecture and depth, not just from having shown up first.
What’s been the biggest operational or strategic challenge you’ve faced at NeuralTrust so far, and how did you pivot or adapt to overcome it?
Scaling from a technical team with a strong thesis into an organization that could serve regulated enterprises like banks and airlines, where procurement cycles, compliance requirements, and analyst validation matter as much as the product itself, was the real inflection point. That meant building out functions that didn’t exist on day one: dedicated security research, formal analyst relations work with Gartner, KuppingerCole, and MarketsandMarkets, and a go-to-market motion built for enterprises with long sales cycles rather than fast self-serve adoption. The adaptation wasn’t a single pivot so much as recognizing early that this category would be won on trust and validation as much as on technology, and building the organization to earn both in parallel.
Where do you see NeuralTrust in the next 18-24 months? What are the key milestones you’re focused on hitting before your next funding round?
The near-term goal is to become the clear market leader in Europe by the end of this year and build from there internationally, particularly deepening our presence in the US through New York. Concretely, that means continuing to expand the product suite’s coverage across the AI agent lifecycle, deepening integrations across TrustGate, TrustGuard, TrustLens, and TrustTest so customers get a single governed layer rather than point tools, and continuing to build the research credibility that got us here, more original threat identification, more analyst leadership positions, more regulated-industry customers who can act as reference accounts as we go into a Series A conversation.
For aspiring founders in the AI or cybersecurity space, what’s your advice on getting incorporation, equity planning, and fundraising readiness right from day one?
Get the boring things right early, incorporation, vesting, equity structure, so you never have to think about them again under pressure. But if I had to give one piece of advice that matters more than any of that, it’s about people. In a category this technical and this new, your first ten hires define what the company is capable of long before any process or playbook does. I’d rather have someone with real conviction about the problem and the humility to learn fast than someone with a perfect resume and no skin in the game. Hire for people who can operate with almost no structure, because in the first year there isn’t any, and you need people who build the structure themselves rather than waiting for it. I’d also say don’t hire too far ahead of what you actually understand about the problem yet. It’s tempting to bring in senior people early to look credible to investors, but if you don’t yet have the internal clarity to direct them well, you end up with expensive people building the wrong thing confidently. Get the founding teams thinking sharp first, then hire people who can execute against that with speed.
