OpenAI tightens safety precautions for video generator Sora 2
openai sora
| Source: Mastodon | Original article
OpenAI has rolled out a new set of security safeguards for Sora 2, its AI‑powered video generator that is embedded in the premium ChatGPT offering. The company announced that every video produced by Sora 2 will now carry both visible and invisible provenance markers, embedding C2PA metadata that identifies the source model, the user account and a cryptographic hash. Access to the model is also restricted to verified enterprise accounts and to individual users who have completed a mandatory “deep‑fake awareness” tutorial. Attempts to generate content that violates OpenAI’s policy – such as realistic depictions of non‑consensual sexual activity or political figures in false contexts – will be blocked by an on‑the‑fly content filter that cross‑checks prompts against a continuously updated risk database.
The move tightens the framework OpenAI first outlined when it launched Sora in late 2025, a tool that promised to democratise video creation by turning short text prompts into fully rendered clips. While the technology opened fresh creative avenues for marketers, educators and indie filmmakers, it also sparked alarm among regulators and civil‑society groups over the potential for mass‑produced deepfakes. By embedding traceable signatures directly into the media file, OpenAI hopes to give platforms and investigators a reliable way to flag synthetic content, a step that could shape future legislation on AI‑generated media.
Watchers will be looking at how quickly third‑party platforms adopt the C2PA standard and whether the provenance data can be spoofed. Analysts are also monitoring OpenAI’s dialogue with European data‑protection authorities, which may influence the rollout of similar safeguards for other generative models. The next test will be whether the stricter gatekeeping slows adoption among creators or proves enough to allay the deep‑fake backlash that has shadowed Sora since its debut. As we reported in September 2025, OpenAI built Sora with security as a foundation; the current upgrade marks the first major iteration of that promise.
Sources
Back to AIPULSEN