Upgrading the Moderation API with our new multimodal moderation model

Source: OpenAI Date: 2024-09-26 URL: https://openai.com/index/upgrading-the-moderation-api-with-our-new-multimodal-moderation-model

Summary

Summary

OpenAI upgraded its Moderation API with a new multimodal moderation model in September 2024 — extending content classification from text-only to text-plus-image inputs. The upgrade enabled developers to moderate both the textual and visual content of user submissions through a single API call, with improved accuracy and category coverage.

Implications

Platform/safety thread. Multimodal moderation is a natural evolution of the Moderation API as GPT-4o made image inputs common in API deployments. The previous text-only moderation was a gap for any application accepting image uploads, where harmful content could be submitted as images to bypass text-based filters. The upgrade also reflects the maturation of OpenAI’s platform strategy: developer tools like the Moderation API reduce the safety implementation burden on API customers, making compliant deployment easier. Each platform safety tool also reduces OpenAI’s liability exposure when API customers build applications that serve users — the “we provided the tools” defense requires the tools to actually be available and capable.