OpenAI’s new AI video-sharing app, Sora, which allows users to generate videos featuring themselves in scenes with popular characters, has sparked copyright concerns. Initial news reports suggested that the company was taking an opt-out approach to copyright, expecting copyright owners to explicitly prohibit the use of their work on Sora. However, Sam Altman, OpenAI's CEO, announced an about-turn, aligning Sora’s copyright policy more closely with the “opt-in model” that it has in place for people’s likeness.  He stated that copyright owners will be given “more granular control” over how their characters are used on the platform, aligning Sora’s policies closer to an opt-in model. Altman emphasized that rightsholders will decide how their content is used and that the company plans to explore revenue sharing with willing creators.

The company still has much to decide relating to copyright and profit sharing on Sora. In the interim, the app – launched so far on an invite-basis to consumers in the United States and Canada – may present legal risks for the company. Liz Corradino notes the potential legal risks: "OpenAI may face claims if the model generates output that is substantially similar to copyrighted works or if the training process itself involved unauthorised use of copyrighted materials." She says that she would expect a company like OpenAI to have several policies and safeguards in place to prevent infringement. These might include:

• content filtering/detection measures to prevent the generation of infringing content;
• careful selection and documentation of AI training data;
• user guidelines and terms of service which clearly prohibit the use of Sora for infringing activities, with explicit consequences for policy violations;
• reporting mechanisms for suspected infringement, with prompt review and takedown mechanisms; and
• an ongoing review and update of these or other policies

OpenAI has implemented safeguards including content filtering, takedown mechanisms, and clear user guidelines, though how granular control will function in practice is still being developed.