Preference Model

Member of Technical Staff - Cybersecurity Capabilities

San Francisco · Staff+

Sponsorship not specifiedDetected 70 days ago

PythonRustC++Machine LearningLLMsCybersecurityResearch

About the role

As part of our goal to automate every role at a hypothetical AI research lab.
One important capability we care about is models' understanding of cybersecurity.
You'll join a small, high-ownership team and contribute directly to the data layer that powers frontier LLM capability in security.

Design and build RL environments and reward functions that produce clean, learnable signals for frontier models on offensive and defensive security tasks across diverse programming languages.
Build environments covering the full vulnerability lifecycle: discovery in source code, exploiting, patching.
Build environments for reverse engineering tasks across binaries, bytecode, and obfuscated code.
Collaborate with others to brainstorm and create new ideas and tools to improve the environment building process.
Problem solvers who take ownership and drive solutions end-to-end.
Visa sponsorship & relocation support available
Experience building or contributing to fuzzing infrastructure, vulnerability scanners, or automated program analysis tools.

Published security research, CVEs, or notable bug bounty findings.
Strong CTF background or competitive results at events like DEF CON CTF, or similar.
Deep expertise in a specific area: binary exploitation, kernel security, browser/V8 internals, hypervisor security, cryptographic implementation, web application security, or cloud/container security.
Experience with ML for code or security.
You have built complex interactive RL environments, agent harnesses, or sandboxed evaluation infrastructure.
Ownership and autonomy in a fast moving startup environment
Weekly snack orders
We value diverse perspectives and experiences.

Preference Model is building automated ML research engineering.
Existing frontier models are brittle when applied to real-world ML tasks. The present bottleneck is the lack of high-quality RL training environments. Our first step is to build RL environments that reflect real-world complexity, with diverse tasks and robust reward functions.
Our founding team has previous experience on Anthropic's data team building data infrastructure, and datasets behind Claude. We are partnering with leading AI labs to push AI closer to achieving its transformative potential.
Existing frontier models are brittle when applied to real-world ML tasks.
The present bottleneck is the lack of high-quality RL training environments.
Our first step is to build RL environments that reflect real-world complexity, with diverse tasks and robust reward functions.
Our founding team has previous experience on Anthropic's data team building data infrastructure, and datasets behind Claude.
We are partnering with leading AI labs to push AI closer to achieving its transformative potential.

This listing is sourced directly from Preference Model's careers page and normalized into a canonical job model.