OpenAI Outlines Principles and Safety Measures for Sora Feed

OpenAI has detailed its philosophy behind the Sora feed, aiming to create a space that fosters creativity and active participation. The core principles include optimizing for creativity through ranking algorithms that favor original content, putting users in control with steerable ranking and parental controls, and prioritizing connections by favoring content that strengthens relationships. Sora's recommendation algorithms will personalize feeds based on user activity, ChatGPT data (with user controls for opting out), content engagement signals, author signals, and safety signals. This personalized system aims to inspire users with content tailored to their unique interests. Safety is a paramount concern, with a multi-layered approach. Strong guardrails at the point of content creation will restrict sexual content, graphic violence, extremist propaganda, hate content, and content promoting self-harm. Beyond generation, the feed will filter out harmful or age-inappropriate content, utilizing both automated tools and human review. OpenAI acknowledges the need to balance safety with freedom of expression, aiming for proactive guardrails combined with a reactive “report + takedown” system. The company plans to continuously refine its safety models based on user feedback.