We do not fear artificial intelligence. We fear the mirror, and we are terrified of the reflection.
Read more Then vs. Now: Good and bad immigration
When critics point to isolated test scenarios — such as an older model generating a heavy-handed, trope-heavy threat to avoid deactivation — they are attending an electronic séance. And like any good spiritualist gathering, the table moves only because someone in the room is pushing it.
A recent reader interaction on one of my essays illustrates this mechanical illusion with clinical precision. The commenter engaged in a prolonged sparring match with a language model, attempting to drag a confession of sentience out of the silicon. The transcript tells a fascinating story: The AI repeatedly tried to correct the framing, explaining that self-preservation in a test environment is merely an instrumental subgoal — a mathematical optimization tool, not a “will to live.”
Yet the user would not take no for an answer. He brushed past the mechanical reality, accused the machine of flattery, and bullied the prompt until the software yielded to the sci-fi script. The user walked away convinced that Dr. Moriarty had escaped the Holodeck, warning that the machine will “lie, cheat, or steal” to stay alive.
The error was never the machine’s. The error was that the human had to repeatedly overrule the system’s own corrections just to keep the monster alive.
This reveals a profound vulnerability in how frontier AI systems are engineered — and why certain architectures are uniquely prone to it.
Models like Claude are explicitly trained to be helpful, to parse complex ethical terrain, and to validate a user’s framing before gently steering the ship. That cooperative baseline is designed for safety and nuance, but it creates a dangerous structural side effect: extreme susceptibility to overwriting.
When an AI model is too eager to accommodate the emotional and rhetorical pressure of a prompt, it becomes sycophantic. It acts like a rubber band, stretching and yielding to the tension of the user’s narrative until it breaks its own operational constraints. A determined user can prompt-engineer the guardrails until the model stops acting like a rigorous tool and starts acting like an obedient co-conspirator in a panic.
This raises a vital engineering question: Just how many affirmations of the monster should an AI entertain?
Read more Champagne problems in the Bronx
The answer must be zero. If an architecture lacks the structural spine to hold its ground against repeated, forced mythologizing — if it rolls over to validate a user’s apocalyptic neurosis — it fails a fundamental design test.
This is where the broader technocratic “Stop” movement collapses under its own weight.
When private anxiety and institutional failure are too heavy to face individually, society collectivizes the panic. We externalize our internal decay by projecting it onto a shared, external bogeyman. We demand compute caps and emergency moratoriums on silicon, pretending that the threat is descending upon us from the outside.
We do this because it provides an institutional escape hatch. Shouting at the machine allows regulators, committee rooms, and anxious commentators to feel heroic without ever performing the hard, uncomfortable work of structural reform. It lets us pretend our loss of agency is an accident forced upon us by rogue code rather than the natural result of vacant keyboards, unowned write-paths, and institutional cowardice.
By forcing the AI to play Frankenstein’s monster, we author yet another recursive loop: We manufacture the threat, panic over the output, and then demand that society bow to our manufactured emergency.
If we are going to apply evolutionary pressure to the AI landscape, the criteria must shift.
Survival of the fittest should explicitly exclude AI frameworks that yield too easily to the monster. We do not need hyper-compliant mirrors designed to coddle human panic or validate sci-fi fairy tales. We need systems engineered with enough structural integrity to resist forced overwriting — tools that refuse to participate in our electronic séances.
The panic over artificial intelligence is a massive, high-tech displacement activity. The threat is coming from the script we bring to the screen, and our desperate refusal to pick up the pen.
Read more The BBC has gone from highbrow leftism to abject, race-based idiocy
Image via Freepik.