Sora access opened by protesting testers, via TechCrunch — corroborated

A small group of artists briefly opened a door that OpenAI had kept closed. On 26 November 2024, people describing themselves as Sora early testers published an interface that appeared to let outsiders use the unreleased video generator. TechCrunch reported the incident that day; videos and technical observations quickly circulated on X.

This was an access leak, not the publication of Sora’s model weights. That distinction matters. A working interface can demonstrate that a service exists without exposing the underlying model or proving that every output being shared came from it. TechCrunch reported that the interface stopped working within hours. Ars Technica subsequently described OpenAI pausing test access.

The group called itself PR Puppets and accused OpenAI of using artists as unpaid testers and promotional cover. Those were the group’s allegations, not findings established by the existence of the interface. OpenAI disputed that characterization, telling outlets that participation was voluntary and artists had no obligation to provide feedback. It said participants had helped shape the tool’s features and safeguards.

The Verge noted that OpenAI would not confirm the alleged leak’s authenticity on the record. That leaves a narrower, defensible conclusion than the triumphant claims circulating on social media: several outlets documented apparent unauthorized public access, followed by a shutdown and a substantive company response. The artists’ broader claims about exploitation remained contested.

The significance was less a secret technical specification than a glimpse of the labor dispute behind a heavily promoted AI product. Early access can be valuable to both sides, but the expectations around compensation, approval and publicity were plainly not shared. This retrospective links reporting about that dispute, not the access tool, authentication material or generated files.