Kroda Labs
AI × Security Research

An independent research lab working at the intersection of AI and security.

We build and break things on both sides of that line, and publish what actually works.

Read our researchWork with us
lab activeest. 2026 · independent
Research01

Three questions we keep asking.

Most of our work fits into one of these three. The problems worth writing about usually sit somewhere between them.

/01

Attacking AI

What happens when a model is the target? We probe language models, agents, and the pipelines around them to find the cracks.

/02

AI that attacks

Offense is getting automated. We study how capable models can be used to perform offensive security research.

/03

AI that defends

The same capabilities cut both ways. We study how those models can be used to defend against attacks moving just as fast.

Publications02

What we've published.

2026-07Baselines Before Architecture: Evaluating Coding Agents for Autonomous Penetration TestingOff-the-shelf coding agents already solve much of the XBOW benchmark, so how much of a pentest harness's score is really the harness?Ananda Dhakal · Krish Neupane · Aarjan Chaudharypaper
About03

Kroda is a small independent lab working where AI and security overlap. Most of that ground is still unmapped, which is roughly why we started.

01 / OPENResearch is published, code is released, results are reproducible.
02 / RESPONSIBLEEvery vulnerability we find is disclosed to the vendor first. Always.
03 / SKEPTICALWe test claims, including our own, before we believe them.
Contact04

Something we should look at?

hello@krodalabs.com

Research collaborations, disclosures, press, or just a strange finding you can't explain. We read everything.