OpenAI brings text watermarking to its API — and unlike Anthropic, it’s off by default
OpenAI has brought text watermarking to its API, The New Stack reports. The central detail in the source is the default: the watermarking capability ships turned off, so developers who want it must opt in rather than receiving it automatically. The article contrasts this with Anthropic, stating that OpenAI's approach differs from Anthropic's in that respect. The available source text does not name specific models, API versions, endpoints, or release dates tied to the watermarking feature, and it does not describe how the watermark is embedded or detected. It also does not state whether the capability applies to all API traffic or only to certain generation paths. Because the excerpt is limited, the digest stays with what is explicitly reported: the feature exists in OpenAI's API, it is off by default, and the outlet positions that default as a point of difference from Anthropic. No benchmark figures, pricing, or accuracy claims appear in the source, so none are repeated here.
Developers calling OpenAI's API will not get watermarked output unless they explicitly enable it, unlike the Anthropic comparison the source draws.