ABSTRACT
AI Responsibility is a critical focus in today's tech landscape. Trust and Safety (T&S) teams have traditionally played a crucial role in risk assessment and mitigation, applying their expertise to ensure product experiences align with company values and enhance user trust. While T&S risk assessment often operates at a feature or content level, increasing adoption of AI suggests practitioners may benefit from lessons learned in assessing risks for AI/ML systems across the technical stack in order to address safety at scale. AI Responsibility and T&S have historically been somewhat distinct domains, with the former often embedded in research or engineering teams and the latter positioned differently across companies, but often in business-related functions. However, there is a growing convergence between these areas. Integration points could significantly enhance responsibility practices by leveraging T&S functions which often apply across the entire business. This chapter introduces a new framework that enables practitioners to draw from insights across both domains to make AI products safer. Generative AI introduces new opportunities and challenges for T&S across traditional functions like content moderation, as well as novel safety issues such as jailbreaking or hallucinations. With new opportunities for abuse, as well as new capabilities, teams must develop enhanced safety tools and processes around AI systems, anticipating distinct issues in different stages of development. This chapter explores increasing intersection points between T&S and Responsible AI, describing tools to equip T&S teams and other interested stakeholders for this rapidly evolving landscape.
