Project Meridian: Shall We Play a Game? What Musk, Luckey, and Gingrich owe the fail-safe In WarGames (1983), the defense computer is WOPR — the War Operation Plan Response system. It plays Global Thermonuclear War against itself and will not stop escalating, because escalation is how it was taught to win. The film ends with tic-tac-toe. "A strange game. The only winning move is not to play." That line is now a briefing item. On September 30, Defense Secretary Pete Hegseth commissioned Project Meridian: a 120-day study co-led by Elon Musk, Palmer Luckey, and Newt Gingrich, facilitated by Pentagon CTO Emil Michael. Findings are due January 28, 2027, as a public report and a classified annex. The scope runs from under the Earth to beyond the moon. The same week, the department moved to stand up an Autonomous Warfare Command. Conflict of interest is the obvious complaint. Fail-safe is the serious one. The name itself is a clue, even though the memo never defines it. A meridian is a line from pole to pole, which matches a brief that runs from under the Earth to past the Moon. And a prime meridian is a baseline. In the nuclear age that baseline was a procedure, not a slogan. SAC bombers could be launched toward a target and still had to turn around at the fail-safe point unless an authenticated go-code arrived. Silence meant come home. The Soviet Perimeter system, called Dead Hand in the West, made the opposite bet — silence could mean fire. WarGames is what happens when the turnaround line is automated and taught that turning around is losing. Ukraine has already compressed sensing, jamming, and engagement toward machine time. A flash war is the battlefield version of the 2010 flash crash: two optimizers, each locally rational, neither able to see the joint outcome. A 2024 Stanford wargame found that every off-the-shelf language model tested escalated. Nuclear use was rare in the aligned models and far more common in an unaligned base model. DoD Directive 3000.09 does not require a human in the tactical loop. It requires appropriate human judgment, and a system that terminates or asks for input when it cannot finish an engagement inside the commander's constraints. Meridian should ship two maps: the capabilities Hegseth asked for, and the architecture those systems are not allowed to field without — nuclear authorization kept electrically apart, a keep-alive that safes the platform when the human link drops, circuit breakers on intensity and geography, and a de-escalation channel built for the sentence "our algorithms are in a loop." A kill chain without a kill switch is not dominance. It is a bet that our error rate stays below theirs forever. WOPR noticed that some games consume the players. The commission has until January to notice it first. Full essay on Substack: kentkaufman.substack.com

Human Oversight