OpenAI adds inline moderation scoring to Responses API and Chat Completions, covering inputs and outputs in one call

OpenAI on June 4, 2026 added moderation scoring to both the Responses API and the Chat Completions API, enabling developers to check the safety of prompts and model outputs in a single generation request rather than running a separate moderation pass.

What's new

The new capability is accessed by including a moderation object in the generation request. According to the OpenAI developer changelog: "Pass a moderation object in a generation request to receive moderation results" covering both inputs and outputs simultaneously.

Key specifics:

• Available in both the Responses API and Chat Completions API
• Evaluates the user's input prompt and the model's output in the same request
• Returns moderation results inline, alongside the generation response
• Removes the need for a separate POST /moderations call for developers already using these endpoints

Context

OpenAI has offered standalone moderation via the /moderations endpoint since 2022, giving developers a way to flag potentially harmful content before or after generation. The prior workflow required two separate API calls: one for generation, one for moderation. For teams that run both on every request—standard in safety-sensitive applications—this meant double the latency and more complex orchestration logic.

The update lands alongside other recent safety-layer additions. OpenAI launched Lockdown Mode in June 2026 to block data exfiltration via prompt injection, and added per-minute container session billing in the same period. Together, the moves suggest a push to embed safety and control features directly into API primitives rather than treating them as separate, optional tooling.

Why it matters

For developers building production applications in healthcare, education, legal advice, or financial services, inline moderation reduces both latency and architectural complexity. Rather than writing orchestration code to issue two sequential API calls and merge the results, the moderation signal arrives with the generation itself.

The dual-direction coverage—prompt moderation catches harmful inputs before they influence a response; output moderation catches cases where a benign-seeming prompt elicits harmful content—enables tighter gate logic. Both signals in one call means fewer opportunities for either to be skipped under time pressure or cost constraints.

This is a quality-of-life improvement rather than a headline capability, but for teams running high-volume, safety-sensitive workloads, the operational simplification compounds quickly across millions of requests.