Why having “humans in the loop” in an AI war is an illusion

The buried issue isn't whether humans are present during AI-assisted targeting decisions — it's whether the tempo of modern conflict makes meaningful review physically possible before an action executes. The legal dispute with the Pentagon is the surface; the deeper problem is that 'oversight' has been redefined downward to mean 'a human was notified,' not 'a human could have stopped it.'

This story sits at the center of a cluster of Anthropic-related coverage from mid-April. The Verge's report on Claude Mythos Preview (April 17) framed Anthropic's cybersecurity model as a potential path back into government good graces — but this MIT Technology Review piece complicates that narrative considerably. If Anthropic is simultaneously suing the Pentagon over AI deployment standards and releasing models designed to court government contracts, the company is pulling in two directions at once. The London expansion piece from WIRED (April 16) adds another layer: geographic diversification reads differently if Anthropic is also contesting the terms under which its models get used domestically. The TechCrunch thawing-relations story (April 18) suggests this tension may be resolving toward accommodation, which is exactly what critics of autonomous weapons oversight would flag as the risk.

Watch whether the Pentagon responds to Anthropic's legal challenge by codifying a specific human-review time threshold in its AI use policy within the next 90 days. If it does not, that absence will confirm the oversight framework is being left deliberately vague.