A group of people has managed to gain unauthorized access to Sora, a video generation tool developed by OpenAI, as a form of protest. They are criticizing OpenAI for what they see as dishonesty and for using artists’ work to promote itself without proper compensation.
On Tuesday, this group launched a project on a platform called Hugging Face, which is used by developers to share artificial intelligence tools. They seemed to have linked their project to OpenAI’s Sora API, which is not available to the public yet. By using special access codes that they likely obtained from an early testing phase, the group created a user-friendly interface that allows anyone to make short videos using Sora.
With this interface, users can create 10-second videos in high definition by simply typing a brief description of what they want. When TechCrunch, a tech news website, tried to use it, they found a long waiting list, but many users on the social media platform X were able to share their creations, most of which had OpenAI’s unique logo on them.
However, by 12:01 p.m. Eastern Time, the access to this tool was cut off, likely because OpenAI or Hugging Face revoked it. The group claims that OpenAI temporarily shut down access to Sora for all artists after just three hours of their project being live.
So, why did this group take such drastic action? They argue that OpenAI is pressuring early testers of Sora—people who help test and improve the tool—to promote it positively without fairly compensating them for their contributions. The group, which calls itself “Sora PR Puppets,” pointed out that many artists have been providing unpaid help to OpenAI, a company valued at $150 billion. They believe that the early access program is more about marketing and public relations than about genuine artistic expression and feedback.
Initially, the group did not disclose who its members were, but as time went on, they started to reveal some of their identities in a post on Hugging Face and in a separate petition.
The group also claims that OpenAI is being dishonest about what Sora can really do. They say that every video generated by Sora must be approved by OpenAI before it can be shared widely, and only a select few creators are allowed to have their videos shown publicly.
The group emphasized that they are not against using AI in the arts; in fact, they were invited to be part of the early testing program. What they oppose is how OpenAI has managed the artist program and how Sora is being presented ahead of its potential public launch. They hope that by sharing their concerns, OpenAI will become more transparent and supportive of artists, rather than just using them for promotional purposes.
In response, an OpenAI spokesperson stated that Sora is still in a “research preview” phase and that the company is working to balance creativity with safety measures before making it widely available. They mentioned that many artists have helped shape Sora’s development and that participation in the program is voluntary, meaning artists are not required to provide feedback or use the tool. OpenAI expressed excitement about offering artists free access and plans to continue supporting them through grants and events.
The spokesperson also noted that artists do not have any obligations to OpenAI other than to use the tool responsibly and to keep certain details confidential while it is still being developed. However, they did not explain what they mean by “responsible” use or specify what information they consider confidential.
Read Also: Uber Is Creating A Team Of Gig Workers To Label Data For AI Models
Since its introduction earlier this year, Sora has faced technical challenges as competitors in the video generation field rapidly advance. Adding to the difficulties, one of the key leaders of Sora, Tim Brooks, left OpenAI for Google in early October. In a recent Reddit Q&A, OpenAI’s chief product officer, Kevin Weil, mentioned that Sora’s progress is being slowed down by the need to refine the model, ensure safety, and scale up computing resources. Reports indicate that the original version of Sora took over ten minutes to produce just a one-minute video.
Early versions of Sora also struggled with consistency, requiring users to create many clips before getting one that worked well. For example, filmmaker Patrick Cederberg had to generate hundreds of clips to find a usable one because the model had trouble maintaining consistent styles, characters, and objects throughout the videos.
The leaked version of Sora appears to be a faster, improved variant, based on code that users on X have discovered. This new version seems to offer some style controls and limited customization options.
According to reports, OpenAI has been training Sora using millions of hours of high-quality video clips across various styles and subjects to enhance the quality of the videos it produces.
In addition to technical challenges, OpenAI has lost valuable partnerships to competitors in the video generation space recently. For instance, in September, a company called Runway secured a deal with Lionsgate, the studio behind the “John Wick” films, to develop a custom video model based on Lionsgate’s movie catalog. Shortly after, another company, Stability, which is also working on video generation models, recruited renowned director James Cameron to its board.
Earlier this year, OpenAI was reportedly meeting with filmmakers and Hollywood studios to showcase Sora, and a former CTO even attended the Cannes Film Festival. However, they have not yet announced any collaborations with major film studios.
This article has been updated to include a statement from OpenAI.
