Reducing bias and improving safety in DALL·E 2

OpenAI said feedback from the DALL·E 2 research preview exposed biased and sensitive image outputs.

The company responded by tightening safety measures, including blocking realistic face uploads and public-figure likenesses.

It also improved content filters and refined automated and human review systems before broadening access.

The update highlights how early testing can help reduce harmful outputs as generative AI reaches more users.