Sora AI (Sora 2) is nothing short of amazing for what it can accomplish. The video generative machine can create almost anything you can imagine. You could create a video of Lebron James in a shadowboxing match with Queen Elizabeth II, and this is fun and all, but this amount of potential being given out to every human with a pulse and an internet connection could cause serious harm in the future.
The only saving grace for us is a small, translucent watermark that moves around the screen as the video goes on. However, this is fairly easy to edit/crop out. Also, if you pay $199 for Sora 2 Pro, you are able to remove the watermark in the videos under a couple of loose conditions, such as: Banning generation of political figures, requiring a text prompt, and needing a person generated to be your own self created online avatar, only if you tell it to generate “a person” you can generate specific people.
Though, this doesn’t mean that you can’t generate a hyper-realistic video of some intense schematic and spread misinformation, or of an “undiscovered” sea creature and convince others online of its existence. This is because the guidelines about a cameo only apply to humans. If it is an animal, or a landscape, or another topic along those lines, The A.I. won’t flag it.
Without the watermark, there are still a couple of telltale signs of an AI-generated video: choppy/strange audio, erratic movements, discoloration, letters and numbers that are written wrong, or something changing throughout the video. Like in a scene in which there is a box on the ground, the video will pan around, and then back, and then it’s gone.
But my fear is that in the coming year(s), and potentially even months, OpenAI will update the product or maybe even make a new one altogether that will outdo its predecessor, lacking any core signs we use to identify A.I. It is baffling that A.I. is progressing at the rate at which it has been. If you look at an AI-generated video from 2023, from the most powerful generative machine available, it is so very clear that artificial intelligence was used due to the low quality, e.g. the video of Will Smith eating spaghetti. But if you compare that to a video that you can generate with a free public resource in 2025, there is no question the more recent free video is more realistic.
By 2027, it is not that far off to say it will be the same way, the A.I. will be bettering itself through learning, perhaps even at a more accelerated rate than from 2023 to 2025. I say this because of the rate of improvement I myself have observed since early October.
According to NAHS Digital Media & Communications teacher Tim Walker, “It’s better for our society to try to figure out how to control it and limit it, keep it inside its box. You know the Terminator movie is basically what might happen since it is a learning program.” Walker thinks we are far from anything bad happening.
There is still one glaring issue with A.I. It sometimes will not follow its code or “directions.” For example, Sora has a code to not create any form of nudity, but it has been reported that it will do it anyway occasionally.
But why does it do this? How is it capable of going against its code? The reason behind this is that Sora performs what is called a “triple layer safety check,” where it runs a check on the prompt for anything negative or harmful. This one is the most vague, because it skims your prompt to see if it contains anything bad, but it tends to misinterpret words in the prompt, or not deem it as potentially harmful.
The next check occurs while Sora is generating the video. If it detects any model in the video that could be deemed as harmful or dangerous, it will terminate the prompt. But again, it could miss something malevolent if it doesn’t deem it as harmful.
The final check occurs once the video is done generating. It runs through the completed video one more time to check if there is any harmful or inappropriate content. It will warn you that it can’t send the video. But even this check could fail, causing a 98.40% success rate with sexual content. This means that over 1% of the time it fails, which may not sound like much, but with the thousands upon thousands of videos generated daily, that’s hundreds of videos that could potentially be contaminated.
To answer the question as to why A.I. is actually dangerous to us as a society, there are many reasons. One reason is people in the future could be wrongly convicted of a crime because of “irrefutable evidence” that is A.I. and vice versa. For instance, someone could make a point that a real video is A.I.
“That would obviously have an awful lot of very bad legal ramifications,” Walker said in response to this theory.
Furthermore, Sora 2 allows for piracy of copyrighted material. This could cause a lot of issues entwined with copyright infringement
“But we are starting to step on the toes of intellectual property. Theft of ideas, and they were starting to steal identities, and you know there’s an awful lot of cheating and stuff that goes on with it,” Walker said.
Personally, it feels more worth it to not use A.I. altogether because it has been shown that it could mislead you by telling you incorrect information.
This does mean that in the future, there may be nothing stopping people from believing that a real video is fake and an A.I. video is real. And this may already be happening to a certain extent. A viral video series was spreading around on TikTok where a man from Panama discovered a meteorite in his back yard, and it eventually started “Growing an alien”. Almost immediately, the internet cried A.I. use. However, it came out that no A.I. was involved: it was modeling and special effects. So what does this mean? People are already accusing non-AI videos of being A.I. which is a warning sign, as it shows we as a society do not have the highly-tuned senses required to tell the difference.
“We’re entering a realm where… really the only way to figure it out is to go into it and say, wait, we went too far. But there probably needs to be some kind of way to start to incorporate in society with a little more control,” Walker said.
All in all, I may not be able to force you to avoid A.I. use, but I urge you to be careful. Whenever you see an AI-generated video on TikTok, YouTube, Snapchat, or Instagram, try to compare it in your mind to a similar video that you saw a while ago. Think to yourself, “Is it more realistic than months prior?” After just a couple of weeks of this, you will come to realize the rapid improvements just as I have.
