Salta al contenuto principale


People Are Crashing Out Over Sora 2’s New Guardrails


OpenAI’s Sora 2 platform started just one week ago as an AI-generated copyright infringement free-for-all. Now, people say they’re struggling to generate anything without being hit with a violation error.

Sora, OpenAI’s new social media platform for its Sora 2 image generation model, launched eight days ago. In the first days of the app, users did what they always do with a new tool in their hands: generate endless chaos, in this case images of Spongebob Squarepants in a Nazi uniform and OpenAI CEO Sam Altman shoplifting or throwing Pikachus on the grill.

In little over a week, Sora 2 and OpenAI have caught a lot of heat from journalists like ourselves stress-testing the app, but also, it seems, from rightsholders themselves. Now, Sora 2 refuses to generate all sorts of prompts, including characters that are in the public domain like Steamboat Willie and Winnie the Pooh. “This content may violate our guardrails concerning similarity to third-party content,” the app said when I tried to generate Dracula hanging out in Paris, for example.

When Sora 2 launched, it had an opt-out policy for copyright holders, meaning owners of intellectual property like Nintendo or Disney or any of the many, many massive corporations that own copyrighted characters and designs being directly copied and published on the Sora platform would need to contact OpenAI with instances of infringement to get them removed. Days after launch, and after hundreds of iterations of him grilling Pokemon or saying “I hope Nintendo doesn’t sue us!” flooded his platform, Altman backtracked that choice in a blog post, writing that he’d been listening to “feedback” from rightsholders. “First, we will give rightsholders more granular control over generation of characters, similar to the opt-in model for likeness but with additional controls,” Altman wrote on Saturday.
An error that appears when I tried to use the prompt "Dracula hanging out in Paris." It says: “This content may violate our guardrails concerning similarity to third-party content."
But generating copyrighted characters was a huge part of what people wanted to do on the app, and now that they can’t (and the guardrails are apparently so strict, they’re making it hard to get even non-copyrighted content generated), users are pissed. People started noticing the changes to guardrails on Saturday, immediately after Altman’s blog post. “Did they just change the content policy on Sora 2?” someone asked on the OpenAI subreddit. “Seems like everything now is violating the content policy.” Almost 300 people have replied in that thread so far to complain or crash out about the change. “It's flagging 90% of my requests now. Epic fail.. time to move on,” someone replied.

“Moral policing and leftist ideology are destroying America's AI industry. I've cancelled my OpenAI PLUS subscription,” another replied, implying that copyright law is leftist.

A ton of the videos on Sora right now are of Martin Luther King, Jr. either giving brainrot versions of his iconic “I have a dream” speech and protesting OpenAI’s Sora guardrails. “I have a dream that Sora AI should stop being so strict,” AI MLK says in one video. Another popular prompt is for Bob Ross, who, in most of the videos featuring the deceased artist, is shown protesting getting a copyright violation on his own canvas. If you scroll Sora for even a few seconds today, you will see videos that are primarily about the content moderation on the platform. Immediately after the app launched, many popular videos featured famous characters; now some of the most popular videos are about how people are pissed that they can no longer make videos with those characters.


0:00
/0:09



0:00
/0:09

OpenAI claimed it’s taken “measures” to block depictions of public features except those who consent to be used in the app. “Only you decide who can use your cameo, and you can revoke access at any time.” As Futurism noted earlier this week, Sora 2 has a dead celebrity problem, with “videos of Michael Jackson rapping, for instance, as well as Tupac Shakur hanging out in North Korea and John F. Kennedy rambling about Black Friday deals” all over the platform. Now, people are using public figures, in theory against the platform’s own terms of use, to protest the platform’s terms of use.

Oddly enough, a lot of memes for whining about the guardrails and content violations on Sora right now are using LEGO minifigs — the little LEGO people-shaped figures that are not only a huge part of the brand’s physical toy sets, but also a massively popular movie franchise owned by Universal Pictures — to voice their complaints.


0:00
/0:09

In June, Disney and Universal sued AI generator Midjourney, calling it a "bottomless pit of plagiarism" in the lawsuit, and Warner Bros. Discovery later joined the lawsuit. And in September, Disney, Warner Bros. and Universal sued Chinese image generator Hailuo AI for infringing on its copyright.


OpenAI’s Sora 2 Copyright Infringement Machine Features Nazi SpongeBobs and Criminal Pikachus


Within moments of opening OpenAI’s new AI slop app Sora, I am watching Pikachu steal Poké Balls from a CVS. Then I am watching SpongeBob-as-Hitler give a speech about the “scourge of fish ruining Bikini Bottom.” Then I am watching a title screen for a Nintendo 64 game called “Mario’s Schizophrenia.” I swipe and I swipe and I swipe. Video after video shows Pikachu and South Park’s Cartman doing ASMR; a pixel-perfect scene from the Simpsons that doesn’t actually exist; a fake version of Star Wars, Jurassic Park, or La La Land; Rick and Morty in Minecraft; Rick and Morty in Breath of the Wild; Rick and Morty talking about Sora; Toad from the Mario universe deadlifting; Michael Jackson dancing in a room that seems vaguely Russian; Charizard signing the Declaration of Independence, and Mario and Goku shaking hands. You get the picture.


0:00
/1:33

Sora 2 is the new video generation app/TikTok clone from OpenAI. As AI video generators go, it is immediately impressive in that it is slightly better than the video generators that came before it, just as every AI generator has been slightly better than the one that preceded it. From the get go, the app lets you insert yourself into its AI creations by saying three numbers and filming a short video of yourself looking at the camera, looking left, looking right, looking up, and looking down. It is, as Garbage Day just described it, a “slightly better looking AI slop feed,” which I think is basically correct. Whenever a new tool like this launches, the thing that journalists and users do is probe the guardrails, which is how you get viral images of SpongeBob doing 9/11.


0:00
/1:23

The difference with Sora 2, I think, is that OpenAI, like X’s Grok, has completely given up any pretense that this is anything other than a machine that is trained on other people’s work that it did not pay for, and that can easily recreate that work. I recall a time when Nintendo and the Pokémon Company sued a broke fan for throwing an “unofficial Pokémon” party with free entry at a bar in Seattle, then demanded that fan pay them $5,400 for the poster he used to advertise it. This was the poster:

With the release of Sora 2 it is maddening to remember all of the completely insane copyright lawsuits I’ve written about over the years—some successful, some thrown out, some settled—in which powerful companies like Nintendo, Disney, and Viacom sued powerless people who were often their own fans for minor infractions or use of copyrighted characters that would almost certainly be fair use.


0:00
/1:35

No real consequences of any sort have thus far come for OpenAI, and the company now seems completely disinterested in pretending that it did not train its tools on endless reams of copyrighted material. It is also, of course, tacitly encouraging people to pollute both its app and the broader internet with slop. Nintendo and Disney do not really seem to care that it is now easier than ever to make Elsa and Pikachu have sex or whatever, and that much of our social media ecosystem is now filled with things of that nature. Instagram, YouTube, and to a slightly lesser extent TikTok are already filled with AI slop of anything you could possibly imagine.And now OpenAI has cut out the extra step that required people to download and reupload their videos to social media and has launched its own slop feed, which is, at least for me, only slightly different than what I see daily on my Instagram feed.

The main immediate use of Sora so far appears to be to allow people to generate brainrot of major beloved copyrighted characters, to say nothing of the millions of articles, blogs, books, images, videos, photos, and pieces of art that OpenAI has scraped from people far less powerful than, say, Nintendo. As a reward for this wide scale theft, OpenAI gets a $500 billion valuation. And we get a tool that makes it even easier to flood the internet with slightly better looking bullshit at the low, low cost of nearly all of the intellectual property ever created by our species, the general concept of the nature of truth, the devaluation of art through an endless flooding of the zone, and the knock-on environmental, energy, and negative labor costs of this entire endeavor.