Research Engineer, AI/ML

Our mission is to make autonomous AI agents resilient to emerging threats to their reliability and security.

We are tackling the greatest challenge in AI: bridging the gap between simulation and production. While models are trained to minimize average-case loss on static distributions, production is an adversarial environment. This lack of adversarial robustness is a major bottleneck for the efficacy of AI in high-stakes applications.

Agent security is fundamentally different from traditional cybersecurity. AI vulnerabilities are amorphous and non-deterministic, and the space of adversarial inputs is unbounded. These vulnerabilities are more persistent than those of the past and require dedicated research to solve.

We are a team of AI safety and security researchers laser-focused on addressing this challenge. We have built RL infrastructure at DeepMind, trained state-of-the-art models at Jane Street, NVIDIA, and Cohere, and published leading research at top conferences (NeurIPS, ICLR, ICML, and more). We care deeply about the safety risks of AI agent adoption and began collaborating as part of the Harvard AI Safety Team. We are a well-funded, early-stage startup backed by top investors.

As a Research Engineer, you will be responsible for post-training models for adversarial capabilities using reinforcement learning.

You will work with the entire training pipeline from data generation and environment design to evaluations.

You will tackle engineering challenges involving distributed systems, innovate new methods for data-constrained and low signal-to-noise environments, and make algorithmic improvements for applying trained models for adversarial simulations.