OpenAI removes watermarks from Sora-generated videos, making it harder to distinguish fake videos

On October 11, 2025 (JST), OpenAI, the developer of ChatGPT, removed watermarks from some videos generated by its video generation AI, “Sora 2.”

As of the time of writing, there is no official announcement from OpenAI.

Sora 2 is a video generation AI released by OpenAI, and it can generate short videos from text prompts.

To make it easier to identify videos generated by Sora 2, OpenAI had until now inserted a Sora logo watermark into generated videos.

However, around October 11, 2025 (JST), there have been a flurry of reports from some users that watermarks are no longer being inserted into videos generated by Sora 2.

It appears that when users subscribed to OpenAI’s paid subscription service “ChatGPT Pro” generate videos from text (t2v), watermarks have been removed, including from past videos.

Videos generated from images and text (i2v), and videos generated by users on the lower-tier “ChatGPT Plus” plan, still have watermarks inserted.

Notably, videos generated via the API that OpenAI provides for developers have had no watermarks for some time.

On the internet, amid Sora 2’s explosive popularity, tools to remove watermarks from videos generated by Sora 2 and tools to add Sora 2 watermarks to ordinary videos have been released.

Because watermarks are no longer inserted in some videos generated by Sora 2, it may become harder to distinguish whether a video was generated by AI.

In addition, tools that add watermarks could be used to misrepresent genuine videos as fakes.

In addition to OpenAI’s Sora 2, companies are releasing video generation AIs one after another, such as Google’s Veo 3 and xAI, the parent company of X (Twitter), with Grok Imagine v0.9, and it is thought that judging the authenticity of information will become increasingly difficult.