OpenAI’s **Sora** video creation tool is implementing considerable measures to deter misuse, including stringent restrictions on portraying real individuals. Introduced on Monday during OpenAI’s “12 Days of OpenAI” celebration, Sora provides an array of functionalities for generating and personalizing AI-based videos. However, users soon discovered that there are significant limitations on what the tool can do.

As noted in its [system card](https://openai.com/index/sora-system-card/), most users are not permitted to upload images of actual people for video generation. This feature is confined to a select group engaged in a “Likeness pilot” initiative. OpenAI indicated that this restriction aims to tackle issues surrounding deepfakes and the unauthorized representation of personal likenesses. A representative stressed that OpenAI is vigilantly watching for inappropriate use, promising to eliminate harmful content, take measures against offenders, and enhance its safety protocols based on initial feedback.

### Justification for the Restrictions

Limiting the representation of real individuals in Sora videos is a sensible move to diminish potential dangers. The tool could be misused for malicious intents, such as fabricating non-consensual deepfakes, representing minors, propagating scams, or generating false information. To mitigate these threats, Sora has been designed to deny specific content types based on both text inputs and image submissions.

For example, Sora will thwart efforts to produce NSFW (Not Safe For Work) or NCII (Non-Consensual Intimate Imagery) material, along with lifelike representations of children. Although imaginary images of children are permitted, realistic ones are not. OpenAI has also adopted various safety measures, such as incorporating C2PA metadata in every Sora video, applying a default visible watermark (which can be removed), and utilizing an internal reverse image search to confirm the video’s origins.

### Future Challenges

In spite of these extensive protective measures, uncertainties linger regarding how Sora will react to widespread “stress-testing” by users trying to circumvent its limitations. For the time being, access to Sora is restricted due to [high demand](https://x.com/sama/status/1866332878499623098), leaving numerous users unable to test the tool.

OpenAI’s cautious strategy illustrates the increasing recognition of the ethical and legal dilemmas presented by AI-generated content. While Sora brings thrilling opportunities for video creation, its rigorous safety protocols underscore the necessity of balancing innovation with accountability.