OpenAI CEO Sam Altman recently mentioned in an episode of This Past Weekend with Theo Von podcast that he felt unsettled while testing GPT-5. “It feels very fast,” he said, adding that he felt “very nervous” during some of the sessions. He even said he “sort of felt like it was the Manhattan project.”
While GPT-5 is expected to be significantly faster and smarter than GPT-4, Altman did not focus on what it could do during the podcast. Instead, he went on to talk about what this could mean, and how little has been done to keep its power in check. “There are no adults in the room,” he said, alluding to the lack of regulations for the technology. He also said, “Something about collectively deciding we’re going to live our lives the way AI tells us feels bad and dangerous. It’s a jarring thought coming from someone responsible for building tools that could soon shape everything from education to employment to relationships.”
READ: OpenAI’s Sam Altman warns of AI voice fraud in banking (July 23, 2025)
This isn’t the first time Altman mentioned risks related to advanced developments in artificial intelligence. He had previously said AI could “go quite wrong” and stressed the need for responsible development.
Despite mentioning these concerns from time to time, OpenAI has also opposed AI regulation, saying that could hamper tech companies’ ability to compete with foreign AI innovations. Altman warned at a Senate hearing in May that requiring government approval to release powerful artificial intelligence software would be “disastrous” for the United States’ lead in the technology.
Altman didn’t go into the technical specifics while discussing his concerns about GPT-5. While GPT-5 is set to launch soon, and is expected to be a step up from GPT-4 in its capabilities, it is uncertain what exactly Altman meant with those comments. Some have expressed skepticism at his comparison to the Manhattan project. TechRadar’s Eric Hal Schwartz referred to the Manhattan Project comparison as “another over-the-top sort of analogy.”
READ: OpenAI expands UK footprint with government deal (July 22, 2025)
“Signaling irreversible and potentially catastrophic change and global stakes seems odd as a comparison to a sophisticated auto-complete. Saying they built something they don’t fully understand makes OpenAI seem either reckless or incompetent,” Schwartz said.
Altman had also recently warned the financial industry of a “significant impending fraud crisis,” of artificial intelligence being used to impersonate a person’s voice to bypass security checks and move money. While voiceprinting has been used as an identification method for wealthy clients since nearly a decade ago, Altman said that this method was no longer completely secure since AI can be deployed to create voice clones that are “indistinguishable from reality.”