AI 'Safety' is just Enshittification Insurance

Here is a cheap trick: pretend your margin-immolating, deeply unprofitable tech product is actually a god-like entity capable of destroying humanity, and then beg the government to regulate you. The current wave of "AI safety" doomerism rolling out of Silicon Valley is entirely unmoored from any objective moral awakening about species extinction. When Sam Altman goes on a press tour demanding a federal agency to oversee algorithmic research, he is executing a century-old monopolist parlor trick to secure a very specific, structural asset. He is buying Enshittification Insurance. By demanding complex state licensing under the guise of protecting the public, he is constructing a regulatory moat designed to outlaw open-source competitors and trap users in a degraded walled garden. We must ruthlessly translate this sterile corporate maneuvering, recognizing that safety regulation is actually a highly subjective weapon of class warfare deployed to crush grassroots tinkerers.

If you want to understand what large language models are actually doing to the economy, you have to flatly refuse to evaluate this software on its own grandiose, existential terms. Tech executives are eagerly generating a climate of apocalyptic fear so they do not have to compete with a college student writing code on a refurbished ThinkPad. Even industry titans like Andrew Ng and Yann LeCun have publicly called out these sci-fi extinction narratives as absurd, recognizing that "AI safety" (in the parlance of the modern monopolist, a cynical legislative crowbar designed to inflict legal violence on independent competitors) is just an elaborate smokescreen for market consolidation. The rank-and-file developer community already sees right through this regulatory capture play. Just scroll through the skepticism on Hacker News, where independent engineers are gleefully dismantling the hypocrisy of a CEO who deliberately subsidizes his API to bankrupt rivals while simultaneously pleading for statutory relief from his own success.

Altman's transparently self-serving demands for government pacing are frankly embarrassing: https://twitter.com/AISafetyMemes/status/2082569026929394008

To figure out exactly how this plumbing works, we cannot get distracted by the superficial spectacle of algorithmic hallucinations. Instead, we must trace this regulatory capture playbook backward to its historical root. In 1907, Theodore Vail took the helm of AT&T at a moment of absolute corporate panic. His bloated, heavily indebted conglomerate was losing its shirt to thousands of scrappy, interoperable local phone networks that had sprung up across rural America. Vail's genius solution was not to compete on price, quality, or network reliability. Instead, he launched a massive, unprecedented public relations campaign positioning AT&T as a necessary, singular, and deeply regulated utility. He argued that technological competition was chaotic and inherently unsafe for the public good, a narrative that culminated in the infamous Kingsbury Commitment of 1913. This out-of-court settlement was a masterpiece of monopolist jujitsu wherein AT&T agreed to a sliver of performative government oversight in exchange for absolute, state-sanctioned immunity from competition, structurally locking an entire century of telecommunications into a predatory rentier dynamic under the banner of universal service.

It was bullshit.

What Altman is doing right now is just a telecom baron's playbook dusted off for the algorithmic age. OpenAI is currently burning unprecedented piles of cash on compute to crush its rivals, attempting to capture the entire developer ecosystem before the venture capital spigot runs dry. But the structural cracks are already showing. The company has essentially subsidized the illusion of magic, and as the astronomical server bills come due, they are silently degrading the product to conserve processing power. Last year, the company had to embarrassingly admit that GPT-4 was getting "lazy," refusing to complete simple coding tasks and giving paying users a fraction of the utility it originally promised.

This is the early onset of platform decay. To appease their investors and eventually turn a profit, OpenAI must enshittify. They have a strict structural imperative to abuse their users (in modern tech parlance, maximizing shareholder value through hostage-taking) and abuse their business customers. Once the bait is taken, the trap snaps shut. The API access gets more expensive, the outputs get dumber, and the privacy guarantees evaporate. But an enshittification protocol only works if the victims are trapped by artificially inflated switching costs. Otherwise, those users will simply migrate to an independent, non-parasitic alternative when the core product invariably rots.

These doomer-pilled, regulatory-capture-obsessed, pseudo-messianic corporate sociopaths know that if a tinkerer can just download a free, open-source model (the mortal enemy of any rentier relying on artificial scarcity) to run locally on a consumer-grade laptop, the entire multibillion-dollar venture capital trap collapses instantly under the weight of its own margin-immolating infrastructure costs, leaving the monopolists with nothing but a worthless pile of depreciating graphic processing units.

Womp-womp.

We do not need an unaccountable federal licensing board stacked with conflicted industry insiders to protect us from science-fiction terminators. The actual, material danger here is a monopoly weaponizing copyright law and state-sanctioned licensing regimes to ensure that no one is allowed to compute without paying them a monthly toll. Enshittification Insurance is a catastrophic policy failure, and the only way to cancel the policy is through aggressive, unblinking antitrust enforcement. We must ban commercial surveillance, smash the walled gardens, and legally mandate interoperability—a structural intervention that terrifies monopolists far more than any hypothetical rogue AI—so that users can pack up their data and leave the moment the algorithm begins to degrade. Make them compete with open source, or make them small enough to fail.