Researcher, Safety - Frontier Assurance

Written by the T&S Network from OpenAI's posting.

Frontier Assurance is a safety research function that designs and supports independent assessments of frontier AI models, safeguards, and risks. It coordinates external evaluations and embedded auditing, manages the misalignment disclosure process, and runs safeguard testing so outside experts can scrutinize OpenAI's safety decisions.

• AI misuse and jailbreaks

What you'd do

• Identify consequential safety claims about frontier models and define the evidence needed to test them.
• Support third-party assessments and embedded audits by turning research questions into experiments and analyses.
• Pair with external researchers to supply data, technical documentation, and context on model behavior and limits.
• Design and build secure access to assessment environments with research, engineering, and security teams.
• Run experiments and analyses for evaluators when direct model access is constrained, with reproducible methods.
• Build tools that gather evidence, run evaluations, and log changes across model and safeguard versions.
• Reproduce external findings and investigate discrepancies with internal results.

What it asks for

• Solid coding background and a record of building research tools or evaluation infrastructure.
• Hands-on work with modern machine learning systems, including setting up experiments and reading their results.
• Ability to work in unfamiliar codebases and debug across system boundaries.
• Sound judgment in experimental design, including identifying confounders and gaps in evidence.
• Understanding of or strong interest in frontier AI risks, alignment, and safeguard evaluation.

Our best summary of the posting, and it can be wrong. The employer's posting is what counts: read it in full on OpenAInorth_east before you apply. If we got something wrong, report this listing.