Self-play is the only path forward in cybersecurity
In the next few years, cybersecurity will dramatically transition from relying on human experts to relying on autonomous systems. In conjunction, society will become increasingly reliant on technology. These confounding factors both raise significant safety risks while creating tremendous opportunities. At Incalmo, we are building the next generation of autonomous cybersecurity systems to ensure that this transition is safe and secure.
Incalmo builds infrastructure for training cyber security capabilities. One form of this infrastructure is RL environments that task models with solving realistic security challenges. These environments are high-fidelity, extremely difficult, and verifiable, enabling models to learn and generalize to real-world security tasks.