It's been less than a week since OpenAI dropped a sister app to ChatGPT, launching a wild, AI deepfake fever dream. Sora is the company's brand-new invite-only social media app, named after its AI video generator. Now, some changes are coming that will affect how you can use it.
Sora's most popular feature is called cameo. It lets you upload a video of your likeness that you and your friends can place into any AI-generated scene. It's how I made a video of CEO Sam Altman claiming Gemini is better than ChatGPT, and how CNET smart glasses expert Scott Stein made a scarily realistic version of him reviewing the new Meta AI smart glasses. The advanced model's capabilities have many people worried about the possibility of misinformation and copyright infringement.
Watch this: Scott Stein Tries Out Sora 2 to Review the Meta Ray-Bans 00:13
When you sign up for Sora, you can opt out of letting other people use your likeness. Bigger copyright holders, like movie and game studios, needed to opt out if they didn't want OpenAI's video model to be able to create clips using their copyrighted characters and imagery.
But this simply isn't how copyright law works, according to Robert Rosenberg, an intellectual property lawyer at Moses and Singer LLP.
"The law has been around a long time that copyright attaches to works from the moment that they're created," Rosenberg said. For OpenAI to say you need to opt out "was a complete nonstarter for the copyright community," Rosenberg told me.
OpenAI reversed its stance late last week after Sora's launch. And the company isn't stopping there. OpenAI's head of Sora, Bill Peebles, said on Sunday that there are going to be new cameo restrictions. While you always had the option to block other people from using your likeness, you can now add restricted keywords or scenarios. For example, you can tell Sora not to allow people to use your face and voice in AI political commentary, Peebles said.
The company is also working on making the white watermark more visible when you download videos. Sora's safety guardrails are still leaning toward overmoderating, as is common with new AI media tools, but the company is refining them to "reduce false negatives and catch loopholes."
These changes seem good in theory, but it'll all depend on how they're implemented.
Sora is unique. It's both a social media platform and an AI video generator. Established law, like Section 230 of the Communications Decency Act, says social media companies are platforms and therefore aren't responsible for what content their users create and share on them. At the same time, entertainment giants like Disney, Universal and Warner Bros. have taken AI firms to court because those models allow users to reproduce copyrighted characters without permission. New changes to Sora might help OpenAI get ahead of similar lawsuits.
... continue reading