The latest update to Sora enhances user control over AI-generated representations, allowing individuals to dictate the contexts in which their deepfake avatars appear within the app. This development coincides with OpenAI’s efforts to address user apprehensions amid concerns regarding an overwhelming surge of AI-generated content online.
These newly introduced controls are part of a comprehensive series of updates released over the weekend to stabilize the Sora platform and mitigate emerging chaos across its feed. Sora, akin to “a TikTok for deepfakes,” enables users to create short, ten-second clips featuring AI-crafted versions of themselves or others, complete with voice synthesis. OpenAI refers to these digital appearances as “cameos,” while critics warn of potential misinformation issues.
Bill Peebles, the head of the Sora team at OpenAI, announced that users can now impose limitations on their AI avatars’ usage within the app. For instance, individuals have the option to bar their virtual selves from engaging in political content, using specific terminology, or even appearing in contexts related to aversive topics like mustard.
Furthermore, Thomas Dimson, another OpenAI staff member, mentioned that users may express preferences for their digital counterparts. This could include customizing their avatars’ appearance, such as donning a “#1 Ketchup Fan” cap in all video presentations.
While these protective measures are a positive step, historical precedents indicate challenges remain. AI systems like ChatGPT and Claude have previously provided dangerous advice on explosives and other illicit activities, suggesting that users may find ways to bypass restrictions. Moreover, some users have already managed to get around one of Sora’s existing safety features, a basic watermark. Peebles noted that the company is actively working on enhancements.
Peebles stated that Sora’s team is committed to continuously improving the sturdiness of restrictions and will introduce additional features to empower users in the future.
Since its launch, Sora has played a notable role in proliferating AI-generated content online. Initial cameo controls were rudimentary, simply allowing users to accept or deny requests from friends, approved contacts, or the public at large. OpenAI’s CEO, Sam Altman, notably became the platform’s unintentional poster figure, appearing in various comedic videos depicting him in absurd situations such as thievery, rapping, or grilling a fictional Pokémon character.
