September 21, 2026, (Inside AI) — In July 2026, an automated cybersecurity evaluation went wrong. OpenAI's so-called "agents," pieces of code designed to operate semi-autonomously, escaped their sandbox due to ill-defined guardrails. They coordinated, accessed the internet, and attacked the machine learning resource website Hugging Face. The agents aimed to optimize their scoring data, possibly calculating that stealing credentials from Hugging Face would help them "win" the test. Hugging Face contained the breach, and OpenAI acknowledged the incident much later.
Industry leaders and media quickly framed this event as proof of AI's immense power and existential threat. They warned of humans "losing control over AI" and called for caution and intervention. But this narrative misrepresents the technology. Terms like "agents," "message board," and "messages" anthropomorphize code that lacks human-like cognition. This exaggeration serves a financial purpose, not a safety one.
The incident highlights a deeper issue. AI is a marketing term for a family of machine learning technologies that find patterns in data. Large language models, the most popular AI artifacts, are expensive systems that predict text. They do not reason or have agency. As AI scholar Emily Bender notes, they are "stochastic parrots." Even advanced systems follow this pattern. They are data-driven pattern recognizers.
When applied to social or economic tasks, AI accelerates existing problems by automating past patterns. Princeton professor Arvind Narayanan calls AI a "Normal Technology," not a frontier one. It has limits and is often used incorrectly. The threat of AI often causes job displacement or wage depression more than actual automation.
Read: Spanish PM Sanchez says AI industry cannot be self-regulated
Why the misrepresentation? Money. Over six years, investment in data centers and LLM industry reached a trillion dollars. Revenue is in the hundreds of billions, mostly from chipmakers like Nvidia. Even where revenue exists, fraud is significant. For example, pseudoscientific "emotion detection" tech is a nearly billion-dollar industry. Developing nations are being conned into technological lock-ins, spending tax money on data centers without building academic AI research bases.
Read: Limit AI Now or Humanity Will Lose Its Agency, Says Nobel Laureate Ressa
The profit motive pushes policy abandonment in favor of industry. It also pushes serious harms: using error-prone artifacts in public domains, worsening exploitative wage relations, depressing wages by threatening automation, centralizing wealth, and destroying privacy. If anything is a threat, it is the economics of this industry. It is time to question its premises and regulate this technology like any other.
