Why is Sam Altman afraid of GPT-5?

Is Sam Altman really afraid of GPT-5 or crafting the perfect buildup? His Manhattan Project comparison and talk of “no adults in the room” frame it as a leap in reasoning, multimodal AI, and agent-like behavior.

If you’ve seen the Oppenheimer movie from a couple of years back, you might remember the early genius of the scientist being harnessed into politics.

And then gradually, layer by layer, he becomes embodied in the Manhattan Project, a top-secret US government atomic bomb development program that ultimately led to the tragedies of Hiroshima and Nagasaki.

On a recent podcast called This Past Weekend with Theo Von, Altman compared GPT‑5’s development to that of the Manhattan Project, revealing he has been “looking at it thinking, What have we done?”

Is he really afraid? Or controlling the narrative?

When a CEO publicly says he’s “nervous” about his own product, it creates immediate intrigue and headlines.

If GPT‑5 is so powerful that even its creator is unsettled, then the stage is perfectly set for any safety features OpenAI announces.

It could well be a PR stunt to build hype and position GPT‑5 as a historical moment in the AI race.

By claiming that “there are no adults in the room,” he gives an eerie sense of how the software could spiral out of control, as opposed to building an edgy or vital version of his AI.

One user observed on X, “When even the creators sound scared, are we just watching autopilot and calling it leadership? Someone needs to take responsibility – fast.”

The adroit commenter pointed out the need for intervention on Altman's part, especially if “there are no adults in the room.”

When even the creators sound scared, are we just watching autopilot and calling it leadership? Someone needs to take responsibility—fast.

undefined Andreas B (@AB_DataPioneer) July 29, 2025

What we actually know about GPT‑5

Currently, online rumours point to major leaps in better reasoning, full multimodal support (text, audio, video), and even early “agent-like” behavior.

When GPT-3 was initially launched in 2020, there were calls from within OpenAI that the model was “too dangerous to release,” especially as it could generate fake news and propaganda.

When GPT‑4 launched, Altman hinted it “might be the first step toward AGI,” a framing that drove headlines and shaped expectations even though OpenAI never confirmed specific capabilities.

This pattern suggests that invoking fear has been part of OpenAI’s playbook to build hype and justify tight control over access.