In January 2026, Sam Altman stood in front of cameras and told the world: “I am now confident we know how to build AGI as we have traditionally understood it.”
Three months later, GPT-5 launched. Users immediately said it felt worse than GPT-4. OpenAI panicked and restored GPT-4o within 24 hours. And then things got even uglier.
This is not a story about a product miss. This is a story about a $852 billion company that has been lying to you — and to its investors — about what it can actually build.
The AGI Claim That Quietly Died
OpenAI’s official definition of AGI is a system that can outperform humans at most economically valuable tasks. Sam Altman said they knew how to build it. Then GPT-5 launched, and he immediately clarified: “GPT-5 didn’t meet the coveted AGI threshold.”
So which is it? Either you know how to build AGI or you don’t. You cannot simultaneously “know how to build AGI” and then release a model that isn’t AGI and shrug. That’s not a technical nuance. That’s a lie dressed up in corporate language.
The AGI claim wasn’t made to inform researchers. It was made to justify a $122 billion funding round that pushed OpenAI’s valuation to $852 billion. The timing is not a coincidence.
The Fake Math Problem Scandal
OpenAI’s vice president claimed GPT-5 solved ten of Paul Erdős’s unsolved mathematical problems — one of the most prestigious challenges in mathematics. This claim spread across every tech publication on Earth.
Except it wasn’t true.
Experts reviewed the outputs and found that GPT-5 had found existing literature that contained solutions — it did not solve the problems itself. It was a search engine result dressed up as mathematical genius.
Yann LeCun, Chief AI Scientist at Meta, called it “self-inflicted.” Demis Hassabis of Google DeepMind called it “embarrassing.” These are not random Twitter critics. These are the people who build AI systems for a living, and they were publicly humiliated by what OpenAI claimed.
Why OpenAI Keeps Doing This
Because it works.
Every time OpenAI makes an outrageous claim, the stock market reacts. Investor sentiment shifts. Competitor budgets get redirected. The hype itself is the product. OpenAI doesn’t just sell software — it sells the narrative that AGI is imminent, and that missing the OpenAI train means missing the future.
Gary Marcus, cognitive scientist and long-time AI critic, called GPT-5 “overdue, overhyped, and underwhelming.” He’s been saying this for years. Every year, the hype machine says he’s wrong. Every year, he ends up being right.
What GPT-5 Actually Is
GPT-5 is a genuinely impressive language model. It writes better than GPT-4. It reasons more consistently. It handles more complex instructions. If it had launched without the AGI framing, without the fake math claims, without the $852 billion valuation riding on it — people would have been impressed.
Instead, it launched as the model that was supposed to change civilization. And it couldn’t even outperform its own predecessor on user preference tests without hand-holding.
The Real Question Nobody Is Asking
If OpenAI is this willing to stretch the truth about what their model can do, what else are they stretching the truth about?
Their safety claims? Their alignment research? Their nonprofit-to-capped-profit structure that somehow justified a near-trillion dollar valuation?
The GPT-5 disaster is not just a product launch that went badly. It is a window into a company that has built its entire business on a story — and the story is starting to crack.
AGI is not coming from OpenAI. Not this year. Probably not next year. And the sooner investors, governments, and users understand that, the better.