SafetyKit scales risk agents with OpenAI’s most capable models

Source: OpenAI Date: 2025-09-09 URL: https://openai.com/index/safetykit

Summary

Summary

A case study (URL slug: safetykit) covering SafetyKit’s use of OpenAI’s frontier models to scale trust and safety risk detection agents — building AI-powered content moderation and risk assessment systems that can process large volumes of user-generated content for platforms that need automated safety enforcement at scale.

Implications

Safety/platform thread. SafetyKit is an interesting case study because it uses AI to address the AI-generated content moderation problem: as AI tools make it easier to generate harmful or violating content at scale, the only way to maintain platform safety at matching scale is with AI-powered moderation. The use of OpenAI’s “most capable models” for safety work reflects the recognition that moderation quality improves with model capability — detecting sophisticated misuse requires models that can understand sophistication. This also represents a commercial opportunity: the AI-generated content problem is creating demand for AI safety tooling companies, and SafetyKit’s model is to provide that infrastructure rather than each platform building it themselves.