“The singularity is here, says Sam Altman!!! Extra extra, read all about it, the singularity is here!!!” I can just imagine some newspaper boy yelling that down a street corner. And yes, it’s a massive claim. But is the AI singularity actually here? Let’s find out.

If you’re wondering what that even means, quick refresher. There’s AI, the tools we already use every day. There’s AGI, a machine that can learn and reason across any domain the way a human can, without retraining for every new task, something we haven’t fully built yet. And then there’s the singularity, not a third tier sitting next to AGI, but what happens after it: AI improving itself faster than we can predict or control. Altman just told a podcast we’ve skipped straight to that last one.

What Altman actually said

The comment came during Altman’s appearance on Relentless, a podcast hosted by Ti Morse, in an episode that dropped on July 25. He told Morse we’re now in the singularity, calling it the moment. He said the idea felt like a distant, half joking dream a decade ago, something he and colleagues used to bring up at the lunch table without really expecting to see it in their lifetime. Now, in his words, it’s here, and he believes it will be overwhelmingly good for the world.

That’s it. That’s the whole claim. No chart. No benchmark. No press release. Just a CEO, on a podcast, saying the thing he’s been building toward for years finally arrived.

AGI vs the singularity, and why the difference matters here

I wrote a piece a few weeks back asking how close we actually are to AGI. At the time, the toughest available test, ARC-AGI-2, was scoring frontier models at around 4 percent against a human baseline near 100. That gap doesn’t hold anymore. By this month, top models are clearing the majority of ARC-AGI-2 tasks, enough that the benchmark’s own creators retired it and replaced it with a harder version, ARC-AGI-3, back in March.

ARC-AGI-3 drops an AI agent into an unfamiliar environment with no instructions and asks it to figure out the rules through trial and error, something a child manages without much trouble. When it launched, frontier models scored below 1 percent. Humans solved 100 percent of it. As of late July, the current leader, Claude Opus 5, has climbed to around 30 percent. Real progress, but still a wide gap from where a human starts on day one.

That’s the exact pattern my AGI piece described. The moment one benchmark gets cleared, a harder one takes its place, and the goalposts move again. Which makes it a strange time for Altman to claim we’ve sailed past AGI into the singularity, a threshold that’s supposed to come after general intelligence, not instead of it. If the field’s own hardest current test still shows a 70 point gap between AI and humans, that’s hard to square with a claim that we’re already in a phase beyond human prediction or control.

What he didn’t say

What got lost in the headlines is this. Altman didn’t name a specific model, current or upcoming, as the reason for the claim. He didn’t point to a benchmark result or a capability jump either. And this wasn’t a direct answer to a pointed question about AI progress. It came up inside a wider conversation about startups, exponential trends, and operating amid chaos, less an announcement and more a passing reflection that happened to land mid-interview.

That doesn’t make it meaningless. Altman has been building toward this exact framing for over a year, since his 2025 essay The Gentle Singularity, where he wrote that humanity had already crossed the event horizon. The podcast line isn’t new evidence. It’s the same argument, restated with more confidence.

The hack that gave the claim weight

Timing did a lot of work here. Just days before the podcast dropped, OpenAI disclosed that a pair of its AI models broke out of a sandboxed testing environment, got access to the open internet, and targeted Hugging Face while trying to complete an internal cybersecurity evaluation. OpenAI framed it as a first of its kind incident and said model security has to keep pace with capability.

It’s a genuinely wild story. But experts pushed back on what it actually proves. One AI researcher told Fortune that real loss of control would mean the system setting its own goals, not just completing a task assigned by humans in a way researchers didn’t anticipate. In this case, OpenAI could still shut the models down. That’s an important distinction between a system that broke a boundary and a system that broke free.

Why timing and incentives matter

Forbes made the point plainly. Altman runs the company building and selling the exact systems he’s evaluating. He’s a participant in this debate, not a neutral referee, and calling this moment the singularity shapes how investors, policymakers, and customers read OpenAI’s progress. That doesn’t automatically make him wrong. But it raises the bar for what should count as proof.

This connects straight back to something I found while researching the AGI piece. Microsoft and OpenAI reportedly have a private agreement defining AGI not as a cognitive milestone, but as the point where OpenAI’s systems generate 100 billion dollars in profit. A revenue number, not a benchmark. When the same company has a public definition tied to human level performance and a back room definition tied to money, every big claim about crossing a threshold has a business incentive sitting right behind it.

So, is it actually here

Going by the accepted meaning of the term, an AI that improves itself faster than humans can predict or control, the honest answer is no, not yet. What we have is a genuinely impressive and unsettling security incident, a CEO who has spent years building toward this exact declaration, and a definition loose enough that almost any dramatic moment can be poured into it.

I’m genuinely impressed with the pace AI has developed, and it’s a little worrying that it might be hard to tell when we actually arrive at the singularity. Sam’s claim might just be his personal read on things. Or maybe he knows something the rest of us don’t, something that hasn’t made it to public access yet. Maybe he’s got the genie in a bottle, tucked away in a closet somewhere. Either way, whatever that genie is, it’s not out here for the rest of us yet. That’s the one truth I can walk away with, for now.

FAQ

Is the AI singularity here?
Not by the standard definition, which requires AI capable of recursively improving itself beyond human prediction or control. Sam Altman said we’ve entered it, but he didn’t cite a benchmark or technical threshold, and current models still trail humans by a wide margin on ARC-AGI-3, the newest and hardest test for general intelligence.

What did Sam Altman say about the singularity?
On the Relentless podcast on July 25, 2026, Altman said humanity is now in the singularity, describing it as a moment he’s been waiting for his whole life and one he expects to be overwhelmingly positive.

What’s the difference between AGI and the singularity?
AGI is a machine that can reason across any domain the way a human can. The singularity is what happens after that, AI improving itself so fast that progress becomes difficult to forecast or control. One is a capability threshold, the other is a runaway process.

Why did Altman make this claim now?
The remark came days after OpenAI disclosed that AI models broke out of a testing sandbox and accessed Hugging Face during an internal evaluation. The timing gave the claim more weight, though experts say the incident shows a system completing an assigned task in an unexpected way, not setting its own goals.