Publications & Appearances

Papers, patents, bylines, podcasts, press, and talks - mostly on AI agents, security operations, and where the two collide.

Research & papers

Cyber Defense Benchmark: Agentic Threat Hunting Evaluation for LLMs in SecOps

arXiv preprint · April 2026 · lead author, 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 recorded from real attack techniques, with no alert and no guiding questions, and have to find the malicious events themselves. Passing takes 50% recall on every ATT&CK tactic, and no model has passed yet. The design is what makes hunting checkable: no one can prove a log is clean, but anyone can check whether an agent found the attacks known to be in it.

Patents

Graph-based artificial intelligence agent

U.S. Patent 12,783,687 · issued November 2026 · co-inventor · assigned to Simbian

An AI agent for investigating security alerts. When a security tool raises an alert, the agent builds an interactive graph with the alert at its center and chains of investigation hypotheses, drawn from a knowledge base, branching off it to guide the investigation. Co-inventors: Prince Saroj, Jonathan Todd, and Ambuj Kumar.

Systems and methods for handling security threats

U.S. Patent 12,530,467 · issued January 2026 · co-inventor · assigned to Simbian

A system that takes a security request written in plain language, parses it into structured data, assembles a workflow from a sequence of templated security actions, and carries it through to resolution. Co-inventors: Ambuj Kumar, Sumedh Barde, John Chong Min Tan, Vishwas Pai, Prince Saroj, and Hardik Maheshwari.

Articles & bylines

Podcasts & talks

AI Hackers, Alert Fatigue, and Cyber Defense

CTO Studio · October 2026 · interview with Josh Rubin · 29 min

Why AI learns offense faster than defense: reinforcement learning environments are easy to build for attacking but nearly impossible to standardize for defending, because every organization is different. We also get into why detection is mostly solved and the real failure is investigation capacity, when an agent should act on its own versus escalate to a human, and the path from Twitter and Afterpay to founding a company.

Press & commentary