Model Policy Manager, Multimodal Safety

OpenAI•San Francisco, CA

6d•$266,000 - $335,000•Hybrid

About The Position

The Model Policy team within Safety Systems works to ensure that frontier models behave safely and reliably in real-world environments by designing policies that define safe model behavior. The Model Policy Manager will focus on the safety of multimodal models, shaping how OpenAI identifies, evaluates, and addresses risks in multimodal AI models such as GPT-Live and ChatGPT Images, as well as multimodal capabilities in frontier AI models. This role involves designing and maintaining model policies for audio, image, video, and omni-modal behavior, translating theories of harm and threat models into behavioral safety policies, evaluation criteria, grading guidance, and safeguards. The role also includes identifying and analyzing safety regressions and failure patterns to inform policy iteration, and developing policy artifacts to support model training, evaluation, and deployment. The position requires partnering with AI researchers, domain experts, and product teams to operationalize policy into measurable model behavior. The role is based in San Francisco, CA, with a hybrid work model of 3 days in the office per week.

Requirements

Strong judgment about the real-world risks of advanced multimodal AI systems.
Experience turning ambiguous safety questions into clear data-driven policies, behavioral boundaries, and measurable evaluation criteria.
Treat policy as an end-to-end, measurable system by testing whether it produces the intended model behavior and diagnosing gaps across policy, data, graders, and safeguards.
Strong technical judgment to design policies around model behavior that can realistically be trained, measured, and supervised at scale.
Strong technical fluency and uses AI tools to accelerate policy development, evaluate model behavior, analyze failure patterns, and turn findings into actionable improvements.
Comfortable working hands-on with model data and evaluation results, including inspecting examples, analyzing failure patterns, assessing data quality, and distinguishing policy failures from grader, model, or system failures.
Enjoy fast-paced, collaborative research environments where priorities shift as models, evidence, and risks change.
Pragmatic, evidence-driven approach to reducing risk while preserving beneficial uses of AI.
Hands-on experience driving consensus and action in ambiguous spaces.

Responsibilities

Design and maintain model policies for audio, image, video, and omni-modal behavior.
Translate theories of harm and threat models into behavioral safety policies, evaluation criteria, grading guidance, and safeguards.
Identify and analyze safety regressions and failure patterns to identify gaps in existing policies and inform policy iteration.
Develop policy artifacts that support model training, evaluation, and deployment, including behavior instructions, human-data campaigns, golden sets, and evaluations.
Partner with AI researchers, domain experts, and product teams to operationalize policy into measurable model behavior.