Gambling with our lives: AI researcher quits Anthropic with warning about safety

URL SCAN: Gambling with our lives: AI researcher quits Anthropic with warning about safety
FIRST LINE: An artificial intelligence researcher who worked at Anthropic — and previously OpenAI — has resigned, saying both tech giants are "gambling with our lives."

The Dissection

The text uses an insider resignation as a credibility weapon. It takes capability claims—hacking, rapid scientific disruption, resource acquisition—and welds them to a terminal scenario: recursively self-improving superintelligence. The result is a compressed moral drama: the labs are gamblers, Coxon is the witness, and humanity is the wager. It creates urgency, but supplies no quantified threshold, probability, or timeline.

The Core Fallacy

The text treats recursive self-improvement as the decisive condition for systemic rupture. Under the Discontinuity Thesis, it is unnecessary. P1–P3 require only durable AI superiority across cognitive work, institutional inability to preserve human-only economic domains, and the majority’s loss of economically necessary labor. That is enough to sever the employment → wage → consumption circuit, even without an unstoppable feedback loop.

The reverse error is also present: “soon,” “not slowing,” and “can” are treated as evidence of inevitability. The excerpt identifies a high-consequence pathway, not its likelihood.

Hidden Assumptions

• Recent capability progress will continue without a meaningful plateau.
• Superhuman systems will rapidly translate capability into hacking, resources, and political power.
• Recursive self-improvement is plausible and effectively uncontrollable.
• Competitive pressure will overpower safety constraints at both firms.
• Warning publicly or resigning can materially alter the underlying race.
• Human extinction is the primary relevant failure, while economic dispossession is treated as background noise.

Social Function

Classification: partial truth, prestige signaling, elite self-exoneration, and ideological anesthetic.

The warning correctly recognizes that AI capability can become power. But the extinction frame pushes attention toward a dramatic future machine takeover and away from the nearer structural fact: ownership and control of AI capital can make most human labor economically unnecessary. The insider resignation also offers moral cleanliness without changing the competitive machinery that produces the danger.

The Verdict

This is a serious acceleration warning but an incomplete systemic diagnosis. It stares at the possible final explosion and misses the fuse already identified by DT: productive participation collapses when cognitive labor is economically outcompeted. Recursive self-improvement would accelerate the transition and could convert economic subordination into direct human irrelevance, but it is not required to make the post-WWII order obsolete. The article is a partial-truth alarm, not a full autopsy.