OpenAI officially launched its highly anticipated video-generation tool, Sora, this Monday. However, the company is withholding access to a critical feature—the ability to generate videos using uploaded images or footage of real people—from the majority of its user base to prioritize safety and mitigate the risks of misuse.
Managing the Risks of Deepfakes and Misuse
The decision to limit access is a strategic move to prevent the proliferation of deepfakes and unauthorized content. OpenAI stated in a recent blog post that while artists value the tool for creative expression, the potential for abuse necessitates an incremental rollout. Consequently, only a select “subset” of users currently has access to the feature.
Beyond restricting access, OpenAI is taking further precautions: users are prohibited from sharing generated videos containing clips or images of real people to Sora’s public discovery feed. Rohan Sahai, Sora’s product lead, acknowledged during a recent livestream that the company is adopting a conservative stance to balance innovation with ethical responsibility, admitting, “We might not get it perfect on day one.”
The urgency of these measures is highlighted by the rise in synthetic media risks. Data from the verification service Sumsub indicates that global deepfake fraud incidents surged more than tenfold between 2022 and 2023.
Enhanced Safety and Moderation Protocols
To combat potential harm, Sora includes specialized filters designed to detect if a generated video depicts individuals under the age of 18. If a minor is identified, the system automatically applies a stricter moderation threshold regarding content involving violence, self-harm, or sexual themes.
Furthermore, all videos generated by the platform include metadata compliant with the C2PA technical standard. This allows platforms that support the standard to verify the video’s origin, helping to distinguish AI-generated content from authentic footage.
Addressing Copyright and Creative Integrity
In response to growing tensions with the creative community, OpenAI has implemented “prompt re-writing” to prevent the system from generating videos in the specific style of living artists. This measure aims to address concerns regarding intellectual property, even as legal battles continue between artists and AI developers over training data usage.
Regarding the training of the model, OpenAI maintains that Sora was developed using a “diverse dataset” comprising publicly available information, proprietary data through partnerships, and custom in-house sets. This remains a sensitive subject, particularly following reports that the model may have been trained on YouTube clips, potentially violating the platform’s terms of service.
Advanced Creative Tools for Users
Despite the restrictions, early access users—including video creator Marques Brownlee—have showcased a robust suite of editing capabilities. The platform includes:
- Storyboard Interface: For creating complex sequences of videos.
- Re-mix Tool: Allows for editing existing videos.
- Blend Tool: Merges two videos while preserving core elements of both.
- Loop and Re-cut: Provides granular control for fine-tuning scenes.
Sora is available starting today for ChatGPT Plus and Pro subscribers residing in eligible countries.