AI Safety & Responsibility Policy Manager
Posted Updated
Description
You will own the creation, iteration, maintenance and implementation of some of Runway's content policies across consumer products, enterprise offerings and third-party model integrations. You will balance user safety, creative expression and operational feasibility, translate policies into LLM prompts, run accuracy audits, act as policy owner in red teaming, and maintain golden datasets as the source of truth for moderation accuracy.
Responsibilities
- Own and maintain some of Runway's content policies, balancing user safety, creative expression, and operational feasibility
- Translate policies into LLM prompts and continuously iterate to drive accuracy improvements
- Track shifts in cultural and market norms and new use cases, and update policies accordingly
- Run ongoing accuracy audits to find and resolve key pain points quickly
- Serve as policy owner in the red teaming process, including maintaining automated test banks
- Turn findings from audits and red teaming into clear, prioritized recommendations for policy, product and engineering partners
- Create and maintain golden datasets that serve as the source of truth for moderation accuracy
- Contribute to work across the broader AI Safety & Responsibility team as priorities evolve
Requirements
- Experience in content policy, trust & safety policy, or a closely related field
- Strong understanding of content moderation systems
- Hands-on experience using machine learning systems for content moderation, policy enforcement, or risk assessment
- Excellent written communication skills, including ability to translate nuanced policy positions into clear, functional documentation
- Strong attention to detail and comfort reviewing large volumes of content, some of which may be sensitive or disturbing
- Strong ability to use data/statistics to inform policy decisions
- Ability to act as a self-starter, taking initiative to improve processes and work products
Skills
Sign in to read and join the discussion.