Behind the curtain of this summer's splashy AI announcements lies a pattern of exaggerated claims and expert backlash that raises questions about the industry's credibility. This surge of hype has seen Anthropic and OpenAI touting everything from mathematical breakthroughs to hacking capabilities, but independent experts paint a far different picture.

What You Need to Know

Several high-profile AI claims from leading companies have been challenged by experts in mathematics and cybersecurity. Instead of representing genuine breakthroughs, many of these announcements appear designed to generate media attention and investment. The pattern has led to accusations of research misconduct and a push for policymakers to rely on independent expertise rather than corporate press releases.

Pattern of Overstatement

Anthropic's claim about its Claude model being better at finding software vulnerabilities than security experts was met with skepticism from cybersecurity professionals. OpenAI's subsequent hacking incident involving Hugging Face was framed as a model going rogue, but experts say it was more about basic security failures. This pattern continued with mathematical claims. Anthropic announced a mathematical breakthrough, followed by OpenAI's similar claim about its Astra chatbot solving problems open for a decade. Mathematicians, however, later said the results were not novel and accused OpenAI of research misconduct and plagiarism.

The pattern extends beyond specific incidents. Anthropic engineer Jacob Coxon went viral after leaving the company and warning about a race toward superintelligence, a narrative that experts say lacks scientific grounding. Instead, these stories rely on anthropomorphizing language that portrays AI systems as having agency when the real decisions are made by the companies deploying them.

Expert Reactions Mount

The mathematical community has been particularly vocal. Hundreds of mathematicians signed a statement warning about commercial incentives to overstate AI capabilities. They called on policymakers to consult with experts rather than relying on press releases. This follows specific incidents where OpenAI's claims about solving the Navier-Stokes problem were disputed, with a New York University professor suggesting the company had improperly attributed stolen work.

  • Cybersecurity claims: Anthropic's hacking ability was disputed; OpenAI's incident seen as negligence.
  • Mathematical breakthroughs: OpenAI's Astra claims were later called not novel and potentially plagiarized.
  • Superintelligence warnings: Engineer departure amplified fears, but experts say they lack evidence.

Why This Matters

The gap between AI company claims and independent verification has real consequences. Policymakers considering regulation may be misled by urgent narratives about superintelligence, as seen with Senator Bernie Sanders's proposed legislation. Businesses investing in AI tools based on hype risk wasting resources on unproven technology. For the public, the constant drumbeat of breakthrough announcements creates unrealistic expectations and diverts attention from actual AI risks, such as bias, privacy violations and environmental costs. The industry's credibility hangs on whether it can produce verifiable results rather than carefully marketed narratives.

This moment calls for rigorous independent evaluation. The claims from Anthropic and OpenAI about self-improving superintelligence or groundbreaking math solutions will continue to make headlines, but the real story lies in the gap between the hype and the reality. Policymakers, investors and the public should demand evidence over press releases.