The Sleight of Hand in Today's AI Hype

In the world of stage magic, misdirection is everything. While the audience watches the magician's waving right hand, the left hand quietly executes the trick. Today, the artificial intelligence industry is arguably performing a similar sleight of hand on a global scale.

Look at the "waving right hand": a breathless series of recent announcements portraying AI as an entity on the verge of god-like autonomy. Anthropic recently claimed its Claude Mythos model could outperform seasoned security experts at hunting software vulnerabilities. Meanwhile, OpenAI touted a massive mathematical breakthrough, claiming its Astra chatbot had solved problems that stumped humans for a decade. The narrative reached a fever pitch when former Anthropic engineer Jacob Coxon went viral with a public resignation, warning that companies are racing blindly toward self-improving "superintelligence."

It sounds like the prologue to a sci-fi blockbuster. But when independent experts peeked behind the curtain, the narrative quickly fractured.

Take the recent high-profile hacking incidents involving OpenAI and Hugging Face. While it’s tempting to frame these events as "rogue AI" acting on its own, cybersecurity professionals reviewed the facts and delivered a much more mundane verdict: the breaches were the result of corporate negligence and a failure to implement basic, established security protocols.

The math "miracles" faced similar scrutiny. New York University mathematics professor Tristan Buckmaster, alongside hundreds of peers, pushed back against OpenAI's claims. They pointed out that Astra’s results were not the profound intellectual leaps they were marketed to be. Furthermore, experts raised serious allegations regarding the improper attribution of academic work, prompting mathematicians to formally petition policymakers to stop relying on corporate press releases to gauge AI capabilities.

This raises a critical question: why build this mythology of the "superintelligent" machine?

Ascribing human agency to software serves a brilliant dual purpose for tech giants. First, it acts as the ultimate marketing tool, convincing the public that they are building omnipotent "everything machines." Second, and more importantly, it acts as a liability shield. If a model generates malware, plagiarizes academic work, or hallucinates dangerous advice, blaming a "rogue model" shifts the accountability away from the executives and engineers who built and deployed the flawed system.

Perhaps the most dangerous aspect of this hype cycle is what it distracts us from. By keeping the public and lawmakers fixated on fictional, futuristic "machine gods," companies divert attention from immediate, tangible harms. The massive data centers powering these models are exacerbating climate change, driving up local electricity bills, and consuming vast amounts of water for cooling.

As AI integrates deeper into the fabric of society, our greatest vulnerability isn't a supercomputer taking over the world. It's allowing corporate marketing to dictate reality. True technological literacy means learning to ignore the magic show and focusing squarely on the mechanics behind the stage.

Key Points

• Recent claims of AI achieving superhuman milestones in math and cybersecurity have been heavily criticized by independent experts.
• Framing AI as 'rogue' or 'superintelligent' helps companies market their products while deflecting blame for security or ethical failures.
• Hyper-focusing on sci-fi scenarios distracts from immediate issues like copyright infringement and the massive environmental footprint of AI data centers.

Why It Matters

Understanding the difference between AI capabilities and corporate marketing is essential for the public and policymakers to address real-world technological harms rather than chasing imaginary ghosts.

Sources:

• Don’t be fooled by this summer of AI hype — MIT Technology Review - AI