Breathless claims about AGI and new capabilities fall apart under scrutiny. In April, Anthropic claimed its model Claude Mythos is better at finding software vulnerabilities than most security experts. Then came the Open AI–Hugging Face hacking incident, after which Anthropic and Meta disclosed similar incidents. Anthropic claimed a mathematical breakthrough; Open AI claimed one too. Anthropic engineer Jacob Coxon went viral announcing his departure, saying the companies are "racing straight towards self-improving superintelligence and gambling with our lives." The press covered these breathlessly, repeating anthropomorphizing framings designed to portray software as incipient AGI.

Once experts examine what happened, a different story emerges. Cybersecurity experts say the hacking incidents are about Open AI’s negligence and failure to adopt basic security practices, not models gone rogue. Mathematicians initially stunned by Open AI’s claim that its chatbot Astra solved decade-old problems later realized the results weren’t novel. They accused the company of research misconduct and plagiarism, saying Astra didn’t make a profound intellectual leap. Weeks later, Open AI claimed its own breakthrough. Two days before, Tristan Buckmaster of NYU’s Courant Institute published a statement suggesting Open AI stole others’ work and improperly attributed it.

Claims of incipient superintelligence aren’t based in good science but in transhumanism, eugenics, and wishful thinking. Math and coding are elevated as pinnacles of human achievement and allow verifiable answers, making them easier to tune systems for. Hundreds of mathematicians signed a statement warning of commercial incentives to overstate capabilities and asking policymakers to consult experts rather than press releases. We echo this call.