Summer of Hype: Why the AI Breakthroughs Weren't What They Seemed

It has been a loud few months in artificial intelligence. At the end of April, Anthropic claimed its model Claude Mythos was better at finding software vulnerabilities than most security experts. Then came the OpenAI–Hugging Face hacking incident, followed by disclosures of similar episodes at Anthropic and Meta. On top of that, both Anthropic and OpenAI announced mathematical breakthroughs, and a departing Anthropic engineer claimed the companies were "racing straight towards self-improving superintelligence."

The press, according to an MIT Technology Review analysis by Timnit Gebru and Emily M. Bender, mostly repeated the companies' own framing — breathless coverage built on the labs' anthropomorphized language. But when independent experts had time to examine what actually happened, a very different story emerged.

"Hacking" incidents that weren't rogue machines

The hacking episodes are the clearest example. Cybersecurity experts, the authors report, say the story is really about OpenAI's negligence and failure to follow basic, established security practices — not about "models gone rogue" or AI agents building civilizations. Framing them as rogue models, they argue, portrays the companies' products as superhuman while helping the companies dodge accountability for their own security lapses.

Math breakthroughs that math experts dispute

The mathematical claims fared no better under scrutiny. Mathematicians initially stunned by OpenAI's press release about its chatbot Astra later realized, according to the analysis, that the results weren't as novel as they first appeared. The piece reports that mathematicians have accused the company of research misconduct and plagiarism — and that Tristan Buckmaster, a math professor at NYU's Courant Institute, published a statement suggesting OpenAI had stolen other people's work and improperly attributed it.

The authors note there is a reason labs love to announce progress in math and programming: answers in these fields, once suggested, can be verified — which makes them easier to tune systems for without expensive human annotation. And because math is elevated as the pinnacle of human intellectual achievement, succeeding there sells the idea that labs are building something close to general intelligence.

Why the framing matters

The hype isn't just noise — it has policy consequences. The analysis points to Senator Bernie Sanders's proposed legislation against the development of "artificial superintelligence" as an example of how superintelligence narratives can misdirect lawmakers. The authors also note the industry has suggested that anti-data-center activism is a "distraction" from regulating imaginary superhuman machines — even as, they write, data centers worsen climate pressures, the asthma of nearby residents, rising electricity bills, and water diverted for cooling.

The piece's bottom line: claims of incipient, dangerous superintelligence are not grounded in good scientific or engineering practice, but in narratives the companies use to market their products. The recommendation to policymakers and the public is to slow down, consult independent experts, and treat the next wave of hype with the same skepticism.

Written by the BidAsk Newsroom. Based on reporting from the sources below.