AI Companies Say They're Building Superintelligence. Experts Say It's Marketing

The last few months have felt like a nonstop parade of AI miracles. Anthropic said its AI was better at finding software flaws than human security experts. Both OpenAI and Anthropic revealed their AI tools had been tangled up in hacking incidents. Then both companies claimed their chatbots solved math problems that had stumped researchers for a decade. Each announcement got breathless headlines.

Then the actual experts looked. Cybersecurity specialists said the 'hacking' stories were really about careless security habits at OpenAI, not machines going rogue on their own. Mathematicians who were initially stunned later said the results weren't as new as advertised, and some accused OpenAI of taking credit for other people's work. Two days before OpenAI's big math claim, a New York University professor publicly suggested the company had stolen and mislabeled someone else's research.

Why do these claims always land in math and coding? Because those are the fields where AI answers are easiest to check automatically — no humans needed to grade each one. That makes them perfect marketing showcases, and it lets companies imply they're building machines that can do anything.

Hundreds of mathematicians have signed a statement warning that the tech industry has a strong financial incentive to exaggerate. Their ask is simple: when lawmakers write AI rules, they should talk to real experts, not read press releases. The dramatic language about 'superintelligence' also shifts blame — if a model is a rogue actor, the company that built it isn't responsible. For everyone else, the takeaway is patience: wait a few weeks after the announcement and see what the specialists say.