Launching Sora responsibly

Source: OpenAI Date: 2025-09-30 URL: https://openai.com/index/launching-sora-responsibly

Summary

OpenAI’s responsible launch post for Sora’s general availability, published September 30, 2025 — approximately 20 months after Sora’s February 2024 initial demo. The post covers the safety measures, content policies, watermarking, and usage restrictions in place for Sora’s GA launch. The “responsibly” framing is deliberate: the 20-month gap between demo and GA was partly due to the extensive safety work documented here — the Sora system card, red-teaming, policy development, and content filtering infrastructure.

Implications

20 months of safety work. Sora’s slow path from demo to GA is the most dramatic example of OpenAI’s staged release approach for high-risk generative media. The February 2024 demo generated enormous attention; the September 2025 launch reflects how much work went into making the product viable without enabling large-scale synthetic video misuse. Deepfakes, non-consensual intimate imagery, political manipulation — all required specific policy and technical mitigations before GA.

The responsible launch as a standard. Sora’s “launching responsibly” framing established a post-launch communication pattern that other labs were watching. Whether the safety measures were actually sufficient, or whether they primarily served as liability management, is a live question that the academic community and regulators will assess over 2026.

Thread: Sora and synthetic video. The launch post resolves the thread that started with the February 2024 demo. The subsequent signals in the Sora thread — the Minne Atairu artistic collaboration, the creative community engagement — show how OpenAI positioned Sora’s creative use cases post-launch.

Watch: Whether any of the predicted synthetic video harms materialize at scale in the first 6 months after GA, and whether OpenAI’s content moderation infrastructure proves sufficient to contain them.