Palantir Leverages Project Maven to Embed AI Directly Into the Kill Chain

Palantir’s Maven Smart System (M.S.S.) now runs the entire kill‑chain with a handful of clicks, a fact detailed in Gideon Lewis‑Kraus’s New Yorker piece “How Project Maven Put A.I. Into the Kill Chain.” The platform, which the Pentagon has been buying from Palantir for over a decade, stitches together raw sensor feeds, satellite imagery, and signals intelligence, then hands the fused product to a language model for final decision‑making. In the latest iteration, Anthropic’s Claude is summoned from a drop‑down menu inside the workflow, where it parses the synthesized data and spits out a target recommendation that can be approved and executed with as few as four mouse clicks, according to the New Yorker.

The integration of Claude was not a pre‑planned upgrade. The New Yorker reports that the surprise deployment came during a covert operation to capture Venezuelan President Nicolás Maduro, when Palantir engineers added L.L.M.‑based functionality to Maven “just in time” for the mission. Anthropic’s own executives later reached out to Palantir for clarification, a move that the Trump administration interpreted as a warning that the vendor might be “faithless.” Anthropic later denied that characterization, but the episode cemented the perception of Claude as a “MacGuffin” that drew attention away from Maven’s core automation, writes technology scholar Kevin Baker in the Guardian.

The fallout escalated when Secretary of Defense Pete Hegseth publicly labeled Anthropic a “supply‑chain risk” after the company refused to grant the Pentagon “all lawful uses” of its models, citing concerns over domestic surveillance and autonomous weapons. The ban, however, was delayed because the military still needed Claude for one final strike. Twelve hours after the dispute, the White House launched Operation Epic Fury against Iran, and the New Yorker notes that the first day’s casualties included 175 children at a primary school in Minab. Congressional Democrats subsequently demanded a detailed accounting of AI’s role in the campaign, underscoring the growing political scrutiny of Maven’s automated targeting.

Kevin Baker’s essay, republished in the Guardian, argues that the real story is not Claude but Palantir’s bureaucratic control over the kill chain. He points out that Maven’s L.L.M. layer is merely a veneer; the underlying system has been automating target identification, prioritization, and weapon release for years. The New Yorker’s reporting confirms that the “four‑click” workflow is now a standard operating procedure, effectively collapsing the traditional human‑in‑the‑loop safeguards that once required multiple layers of verification. This shift raises profound ethical questions about accountability when a software platform, rather than a person, selects a target for destruction.

Despite the controversy, Palantir continues to expand Maven’s reach. Bloomberg’s defense‑tech reporter Katrina Manson, cited in the New Yorker article, has been tracking the platform’s evolution and notes that the Pentagon’s reliance on Maven has only deepened as the Department of Defense pushes for faster, AI‑driven decision cycles. The New Yorker’s account suggests that Palantir’s role is now less about providing a data‑fusion tool and more about embedding an end‑to‑end AI pipeline that can move from raw intelligence to kinetic action with minimal human oversight. As the technology matures, the line between software and weaponry blurs, and the debate over who—if anyone—is responsible for the outcomes of that pipeline is only just beginning.