About

I'm Alankrit Chona. I work on a question that sits under most of AI in security: how do you trust an AI with work nobody can easily check? I come at it as a researcher and as a builder.

I studied Computer Science at IIT Delhi, then spent years in founding and early engineering roles - building backend infrastructure at scale at Twitter, and helping build companies including Afterpay and Spotnana from the ground up. Different domains, same underlying question every time: how do you build systems that actually work at scale?

Security is where trusting the work gets hardest. A passing test shows a fix works, but nothing proves a dismissed alert was benign or a quiet hunt was clean. How cheaply work can be checked decides how far AI can be trusted with it, and defense sits near the expensive end. I lay out the argument in The Work You Can't Check.

On the research side, I'm lead author of the Cyber Defense Benchmark, with Igor Kozlov and Ambuj Kumar. It's the first public benchmark for open-ended threat hunting by AI agents: agents get raw Windows event logs from real attack techniques, with no alert and no guiding questions. It swaps a question nobody can check ("is anything wrong here?") for one anyone can ("did you find the attacks we know are there?"). No model has passed it yet.

On the building side, I'm co-founder and CTO of Simbian, where we build AI agents for security operations. I came at cybersecurity from the AI and engineering side, not the other way around, which shapes how I see the whole space.

This site is where I think out loud about AI agents, security, and the things I get wrong along the way.

What I focus on

• AI systems and agents - how they work, where they fail, and what "production-ready" means when the output is hard to check
• Security and verification - why so much of defense resists measurement, and how to test a defense instead of trusting it

For papers, patents, bylines, press, and talks, see publications & appearances.

Get in touch

Find me on LinkedIn.