OpenAI pulled the plug on Sora on April 26, 2026. The text-to-video model and its attached social media app are gone. What remains is a two-year experiment in letting artificial intelligence make moving pictures — and a clear picture of how hard that business turned out to be.
Sora’s first public preview landed in February 2024. The videos were short, strange, and stunning. A woolly mammoth in a snowstorm. A Tokyo street at golden hour that had never been filmed. By December 2024, paying ChatGPT Plus and Pro users in the United States and Canada could make their own. The model could generate clips from scratch or extend an existing video. It was a new tool, and it worked.
Then came Sora 2 in September 2025. This version added social media features. Users could share what they made, comment, scroll. OpenAI was trying to build a community around the technology. It was a pivot from tool to platform. That move, in retrospect, looks like a gamble that did not pay off.
The copyright problem was baked in from the start. Sora’s generator used copyrighted material by default. Copyright holders had to actively opt out to keep their content out of the training data. That is the opposite of how most licensing works. It put OpenAI in a legal posture that was always going to attract lawsuits and regulatory scrutiny. The company knew it. The watermark was their answer — a visible, moving mark embedded in every video. It was supposed to prevent misuse and prove provenance.
It did not work for long. Third-party programs that could strip the watermark appeared within a week of Sora 2’s release. Once the watermark was gone, there was no way to tell a Sora video from real footage. The safeguard became a speed bump, not a wall. That is the kind of failure that kills trust in a product meant for public sharing.
The shutdown means those social features are gone. The video generator is gone. The whole experiment is dark. What comes next depends on what OpenAI does with the underlying technology. The model itself — the thing that could turn a sentence into a moving image — still exists. The company has not said it is abandoning the research. It has said the app is shut down. That is a narrower statement, and it matters.
Building a social network around AI video was always a high-risk play. Social platforms need users, trust, and content moderation at scale. Sora had none of those things locked in. It had a watermark that people could already remove. It had a copyright policy that put the burden on creators to opt out. It had a user base limited to two countries. That is a thin foundation for a platform meant to compete with TikTok or Instagram.
The tech world will watch where the Sora team goes next. The model itself was a breakthrough. The social layer was an experiment that failed. OpenAI may license the video engine to other companies. It may build a professional tool for filmmakers and advertisers. It may just keep the research going in-house. What it will not do is pretend that the social media version worked.
The shutdown is a reset. It is also a reminder that even the most impressive AI demo has to survive the real world — copyright law, security flaws, and the simple question of whether people want to use it. Sora 2 answered that question. The answer was no.
