When OpenAI released its new AI-generated video app, Sora, last week, it launched with an opt-out policy for copyright holders: media companies would have to specifically state that they didn’t want their AI-generated characters all over the app. But after days of Nazi SpongeBob, criminal Pikachu, and Sora Philosophy on Rick and Morty, OpenAI CEO Sam Altman announced that the company would backtrack and “let the rights holders decide how to proceed.”
In response to a question about why OpenAI changed its policy, Altman said it came from his discussions with stakeholders and suggested he didn’t expect an outcry.
“I think the theory of what this was going to feel like for people, and then seeing the thing, people had different responses,” Altman said. “It was more different from the pictures than people expected.”
The TikTok-like Sora app offers endless scrolling and the ability to create 10-second videos — with audio included — of virtually anything, including your own AI-generated self (dubbed “cameo”) and anyone who has given consent for you to use their image. Although it attempts to restrict depictions of people who aren’t on the app, its text prompts have proven more than capable of generating copyrighted characters.
Altman said many rights holders are enthusiastic, but they want “a lot more controls,” adding that Sora “became very popular very quickly…We thought we could slow the ramp down; That didn’t happen.”
“We obviously care a lot about the rights holders and the people,” he said. “We want to create these additional controls, but I think a lot of major content items will be available with restrictions on what they can and can’t do.”
Among early adopters of the system, Altman said he was surprised that people had “middle-of-the-road” feelings about allowing people to create AI-generated videos using their image on Sora. He said he expected people would want to try to make their cameo public or not, but there wouldn’t be as much nuance, and that’s why the company recently introduced more restrictions. Many people have changed their minds about whether or not they want their cameos to be public, Altman said, but “they don’t want their cameo to say offensive things or things that they find deeply problematic.”
Bill Peebles, head of Sora at OpenAI, posted on Sunday
Rights holders want “much more control” over Sora
Peebles also said the team was working on ways to make the Sora watermark on uploaded videos “clearer and more visible.” Many people have expressed concerns about the misinformation crisis that could naturally arise from hyper-realistic AI-generated videos, especially when the watermark denoting them as AI-generated is not very large and can be easily removed, according to the video tutorials that proliferate online.
“I also know that people are already finding ways to remove it,” Altman said of the watermark during Monday’s Q&A.
During Altman’s DevDay keynote, he said the company was immediately releasing a preview of Sora 2 in the OpenAI API, allowing developers to access the same model that powers Sora 2 and create ultra-realistic AI-generated videos for their own needs, seemingly without any sort of watermark. During the Q&A session with reporters, when asked how the company would implement protections for Sora 2 in the API, Altman did not answer the question specifically.
Altman said he was surprised by the scale of demand for generating videos just for group discussions, that is, for sharing with just one other person or a handful of people, but not more broadly than that. While this is popular, he said, “it doesn’t fit perfectly with how the current app works.”
He positioned launch speed bumps as learning opportunities. “We won’t have the one good video model available for much longer, and there will be a ton of videos without any of our safeguards, and that’s fine, that’s the way the world works,” Altman said, adding, “We can use this window to get society to really understand, ‘Hey, the rules of the game have changed, we can generate almost indistinguishable video in some cases now, and you have to be ready for that.'”
Altman said he thinks people don’t pay attention to OpenAI technology when people at the company talk about it, but only when they release it. “We need to have…this kind of technological and societal co-evolution,” Altman said. “I believe it works, and in fact I don’t know of anything else that works. There will clearly be challenges for society with this quality, and what will improve, with the video generation. But the only way we know to help alleviate this problem is to get the world to experience it and understand what it’s going to be like.”
“There will clearly be challenges for society in the face of this quality”
This is a controversial view, especially for an AI CEO. For as long as AI has existed, it has been used in ways that disproportionately affect minorities and vulnerable populations – from wrongful arrests to AI-generated revenge porn. OpenAI has guardrails in place for Sora, but if history – and last week – is any guide, people will find ways around them. Watermark removal tools are already proliferating online, with some people using “magic eraser” tools and others coding their own methods to convincingly remove the watermark. Currently, text prompts do not allow generating a specific face without permission, but people have reportedly circumvented this rule to generate approximations close enough to someone to incite fear or make threats, and to make suggestive videos, including videos of women holding dildo-like objects.
When asked if OpenAI’s plans amount to a “move fast and break things” approach, Altman said, “Not at all,” adding that user criticism of Sora currently tends to be that the company is “way too restrictive” and “censorious.” He said the company was starting the rollout cautiously and would “find ways to allow more over time.”
Altman said making a profit on Sora was “not in my top 10 concerns but… obviously one day we have to be very profitable, and we are confident and patient that we will get there.” He said that at present the company is in an “aggressive investment” phase.
Whatever the initial difficulties, Greg Brockman, president of OpenAI, said he was struck by Sora’s adoption curve — and that it was even more intense than ChatGPT’s. It has remained consistently at the top of the list of free apps in the Apple App Store. “I think it hints a little bit at the future – this thing that we keep coming back to: We’re going to need more computing,” he said. “To some extent, that’s the number one lesson from (Sora’s) launch so far.”
It was essentially a pitch for Stargate, OpenAI’s joint venture with SoftBank and Oracle, to strengthen AI infrastructure in the United States, starting with a $100 billion investment and adding up to $500 billion over four years. President Donald Trump has championed the initiative, and OpenAI has announced a handful of new data center sites in Texas, Ohio and New Mexico. These energy-intensive projects are controversial and often manage to operate with a workforce of a few hundred people after initial construction, despite promises of large-scale job creation.
But OpenAI is moving forward at full speed. On Monday, it struck a deal with chipmaker AMD that could see OpenAI take a 10% stake. Hours later, in a question-and-answer session with reporters, Altman was asked how interested the company was in building its own chip. “We are interested in the whole AI infrastructure,” he replied. At another point, he told reporters that they should “expect to hear a lot more” from OpenAI on the infrastructure stack.
During the session, OpenAI executives repeatedly highlighted the lack of compute and how that can prevent OpenAI and its competitors from offering services at scale.
“Ask: ‘How much computing do you want?’ » It’s a bit like asking: “How much of the workforce do you want?” “, Brockman said. “The answer is you can always get more out of more. “And right now, greater ability to deep-spoof your friends is the latest selling point.
