Researcher, Misalignment Research

San Francisco, United StatesOn-site

Responsibilities

- Create automated tools and infrastructure to scale automated red‑teaming and stress testing.
- Conduct research on failure modes of alignment techniques and propose improvements.

Worst‑Case Demonstrations – Craft compelling, reality‑anchored demos that reveal how AI systems can go wrong.
About the Role We are seeking a Senior Researcher who is passionate about red‑teaming and AI safety.
In this role you will design and execute cutting‑edge attacks, build adversarial evaluations, and advance our understanding of how safety measures can fail—and how to fix them.
We aim to concretely reduce existential AI risk. - Partner with engineering, research, policy, and legal teams to integrate findings into product safeguards and governance processes. - Mentor engineers and researchers, fostering a culture of rigorous, impact‑oriented safety work.
experience in AI red‑teaming, security research, adversarial ML, or related safety fields.
- Are fluent in modern ML / AI techniques and comfortable hacking on large‑scale codebases and evaluation infrastructure.
- Hold a Ph.D., master’s degree, or equivalent
experience in computer science, machine learning, security, or a related discipline (nice to have but not required).
About OpenAI OpenAI is an AI research and deployment company dedicated to ensuring that general-purpose artificial intelligence

YOU MIGHT THRIVE IN THIS ROLE IF YOU - Already are thinking about these problems night and day, and share our mission to build safe, universally beneficial AGI and align with the OpenAI Charter. - Have 4+ years of

ABOUT THE TEAM Safety Systems sits at the forefront of OpenAI’s mission to build and deploy safe AGI, ensuring our most capable models can be released responsibly and for the benefit of society.
Within Safety Systems, we are building a misalignment research team to focus on the most pressing problems for the future of AGI.
Our mandate is to identify, quantify, and understand future AGI misalignment risks far in advance of when they can pose harm.
The work of this research taskforce spans four pillars: 1.
We focus especially on high importance cases where misaligned AGI could pursue goals at odds with human well being. 2.
Adversarial & Frontier Safety Evaluations – Transform those demos into rigorous, repeatable evaluations that measure dangerous capabilities and residual risks.
Topics of interest include deceptive behavior, scheming, reward hacking, deception in reasoning, and power-seeking, along with other related areas. 3.
System‑Level Stress Testing – Build automated infrastructure to probe entire product stacks, assessing end‑to‑end robustness under extreme conditions.
We treat misalignment as an evolving adversary, escalating tests until we find breaking points even as systems continue to improve. 4.
Alignment Stress‑Testing Research – Investigate why mitigations break, publishing insights that shape strategy and next‑generation safeguards.