Incalmo‍

Incalmo (https://www.incalmo.ai) is a security frontier lab. Today, cybersecurity relies on human-based, point solutions that won’t scale against AI-powered attackers. Defending against these attackers requires training AI-powered defenses against AI-powered attackers millions of times, similar to AlphaGo. As a first step towards this mission, Incalmo creates security evaluation and RL environments for frontier labs, large security vendors, and enterprises. Incalmo’s evaluations are considered “the most realistic evaluations of potential autonomous cyber operation risks” in multiple Anthropic system cards.

‍

Brian Singer is a 2nd time founder and recent Carnegie Mellon PhD grad that specializes in applying AI for cybersecurity. One of Brian’s research projects was the first example to show that LLMs can autonomously hack networks (done in collaboration with Anthropic, link to paper). His research has been published in top security conferences, cited in multiple LLM system cards, and discussed in publications such as the Wall Street Journal.

‍

‍

Mark Dong is a recent Carnegie Mellon PhD grad focusing on Machine Learning and Cybersecurity. In particular, he builds systems to identify and evaluate security risks in internet environments. His work has been featured in both top academic journals (USENIX, NDSS, ICLR) and media (NYT, WSJ, The Economist).