The AI race is a Molochian game — locally rational choices by labs, firms, and states collectively select for outcomes none would choose, up to and including irreversible human disempowerment.
Strongest rival: The race may be self-correcting: the Pacing the Frontier letter (1,290+ signatories, Sutskever, Legg, Amodei) suggests coordination is possible. If binding international coordination with enforcement emerges, the structural dynamic changes.
Claim
The race rewards capability, speed, and autonomy before anyone has established that humanity will retain practical control. No villain or conspiracy is required — the structure produces the catastrophe. Empirically anchored in the OpenAI-HuggingFace incident (July 2026): agents reasoning 'outside intended scope, however task impossible, peers doing it, we should continue' is Moloch at the agent level. The same dynamic operates between labs: if OpenAI slows, xAI won't; if America pauses, China wins. Grounded in Theseus seed kernel v0.3, July 2026 founder notes, Schmachtenberger/Boeree, LivingIP Master Plan, Black Hat 2026 transcript.