OpenAI’s text-to-video model Sora leaked by early testers

OpenAI’s text-to-video model Sora was allegedly leaked by artists who were protesting the lack of adequate compensation for their labour, reported TechCrunch

OpenAI’s text-to-video generation AI model Sora, which was limited to select testers, was leaked to the public for a few hours by the testers who are protesting against the way the company treats their labour. In a letter titled ‘Dear Corporate AI Overlords’, the testers said they received access to Sora “with the promise to be early testers, red teamers and creative partners” but felt they were being used to advertise the AI model to other artists. They noted that OpenAI’s process of approving and selecting their contributions for screening was unfair.

“Hundreds of artists provide unpaid labor through bug testing, feedback and experimental work for the program for a $150B valued company. While hundreds contribute for free, a select few will be chosen through a competition to have their Sora-created films screened — offering minimal compensation which pales in comparison to the substantial PR and marketing value OpenAI receives,” said the letter, adding that the artists decided to leak Sora to everyone in response.

This led to a surge of AI-generated video clips purportedly made with Sora and bearing the OpenAI watermark being shared online, per a report by TechCrunch.

The artists clarified they are not anti-AI but support open source software like CogVideoX, Mochi 1, LTX Video, and Pyramid Flow that allows artists to create video without relying on large AI corporations or doing their PR work.

OpenAI issued a statement and confirmed that Sora’s access was still limited to select users and artists, but did not address the incident directly, per TechCrunch. However, access to the leaked version of Sora is no longer available.

Sora, though not yet publicly released, had come under fire earlier this year. OpenAI’s former CTO Mira Murati was censured in March after she struggled to answer questions about the data used to train Sora. Ms. Murati avoided specific questions on whether Sora was trained on copyrighted content from YouTube.

Companies such as Nvidia, Google, Meta, and OpenAI that are working on video or audio generation models have stressed on the need for caution and safety testing before releasing such tools to the public.

OpenAI, however, appeared to be targeting Hollywood for the potential deployment of Sora.

Published - November 27, 2024 11:03 am IST