Ambuj
Kumar

Mid-stride, building something that shouldn't exist yet

The average CISO lasts 18 months before burning out or being replaced. Ambuj Kumar knows this. He's spent the last three years building the reason that might finally change.

Simbian's AI agents - not copilots, not dashboards, actual autonomous agents - now handle the kind of security workload that used to require entire analyst teams cycling through overnight shifts. Threat advisory analysis that previously took hours completes in seconds. Kumar watched a customer's face during a live demo when that clicked. He didn't need a slide deck to explain the value proposition after that.

The thing about Kumar is that he is not a security person who learned AI. He is an engineer who built the infrastructure that makes both possible. Eight years at NVIDIA, as one of the company's first ~500 employees, designing GPU memory controllers under Jen-Hsun Huang. The hardware that now trains the large language models that everyone is scrambling to deploy in their products. Kumar built some of that foundation, then left to figure out what to actually do with secure data at scale.

He has done this twice. The first time, with Fortanix: Confidential Computing, the first commercially successful technology for encrypting data in use - not just at rest or in transit. IBM adopted it. Intel adopted it. Goldman Sachs adopted it. Kumar raised $135 million before stepping back in 2022. The second time, with Simbian: autonomous AI security agents that eliminate 90% or more of routine SOC tasks. He co-founded it with Alankrit Chona, a former Twitter software engineer, and had the seed round oversubscribed before they finished asking.

He grew up in Bihar, India. Earned a Gold Medal at IIT Kanpur. Built an M.S. at Stanford while working full-time at NVIDIA. Holds over 100 patents across cryptography, cloud computing, and computer science. That's a resume that could support a comfortable career as a deep technical executive somewhere. He keeps choosing harder paths.

His philosophy on hiring sounds counterintuitive until you hear the reasoning: "We value hustle, hunger, and the ability to learn over experience. Too much experience can hold you back." When you're building categories that don't yet exist, you need people who don't know what's impossible.

Kumar maintains a strict weekday/weekend split - work entirely on weekdays, family entirely on weekends. For someone running a venture-backed security startup in a market moving at adversarial speed, that kind of discipline reads less like life-balance advice and more like an engineering constraint: the system performs best with defined boundaries and predictable cycles.

The average enterprise manages 76 security tools. More than half of security teams believe they've wasted roughly 50% of their cybersecurity budgets on tools they can't fully use or integrate. The shortage of trained security professionals sits at 3.5 million open positions globally.

Simbian's answer is not another dashboard. It is autonomous AI agents that accept security goals in plain language, reason through the organization's specific context and policies, and execute - integrating with SIEM systems, XDR platforms, and existing tooling rather than replacing them. SOC AI agents that process threat advisories, build threat profiles, scan for environmental compromises, and generate remediation plans. GRC agents that automate responses to security questionnaires. Threat hunt agents that run continuously, not just when someone has bandwidth.

Kumar describes the underlying technology as the security industry's first reasoning LLM specifically trained for security operations. Simbian built what they call the "Context Lake" - a security knowledge graph that gives agents deep domain expertise without hallucinating on edge cases. The agents function as virtual employees, operating under human-defined goals and policies, with humans reviewing outputs rather than doing the work from scratch.

There's a game embedded in the product development process: "Are you Smarter than an LLM" - a gamified platform where hundreds of users test the AI's limitations daily. Crowd-sourced adversarial testing, baked into the feedback loop. This is how Kumar addresses LLM reliability concerns in a domain where false negatives have real consequences.

Gokul Rajaram, Coinbase board member and Simbian investor, framed the team's edge precisely: "I'll put my money on Simbian. It's the team with unique experience all the way from hardware to cloud." That range - GPU architect to cloud security to AI agents - is not a coincidence. It is the reason Simbian's stack runs differently than competitors who arrived at security from only one direction.