Pentagon pivots to AI

The Pentagon is steering more money toward artificial intelligence systems it can field, test, and upgrade now, not just study in labs. (theguardian.com) The Guardian reported on April 22 that the Defense Department requested about $54 billion tied to accelerating AI-enabled warfare in its latest budget plan. The shift favors procurement and integration work over longer-horizon research programs, according to the report. (theguardian.com) In Pentagon terms, that means buying the parts that let existing aircraft, drones, and command systems use AI at the edge — on the platform itself, with a human still in the loop. A January 9 department memo said the goal is an “AI-first warfighting force” built around modular systems and continuous test and evaluation. (defense.gov) The Pentagon has been moving this way for months. A September 27, 2024 memo on Replicator 2 said the department wanted faster fielding, open system architecture, and system integration, with a plan to deliver improved counter-drone protection within 24 months of congressional funding. (defense.gov) One example is the Air Force’s Collaborative Combat Aircraft program, where Collins Aerospace and Shield AI are building mission autonomy software for uncrewed jets that fly alongside piloted fighters. Air Force officials said in February that the software must fit the government’s common autonomy architecture so algorithms can move between vendors’ aircraft. (airandspaceforces.com) RTX said on February 20 that Collins’ Sidekick software completed a four-hour autonomous flight on General Atomics’ YFQ-42A, with a human operator on the ground managing the mission. The company said the test paired the uncrewed aircraft with crewed fighters to extend sensor reach and improve mission performance. (rtx.com) The same buying logic is showing up outside combat programs. Voxelis says its VoxVision system puts AI computing, thermal imaging, and mapping tools directly on helicopters so crews can generate wildfire intelligence in real time instead of waiting for data to be processed after landing. (voxelis.ai) Voxelis has pitched that package as a mass-deployable edge system for civilian helicopters, and launch partners including Contour Helicopters and Custom Helicopters have tied it to wildfire suppression missions. Vertical Magazine’s April feature used it as a case study in how operators want onboard AI that helps crews during the flight, not just analytics after it. (verticalmag.com) The Pentagon is also trying to answer a criticism that faster AI adoption can outrun oversight. A Defense Department responsible artificial intelligence pathway says systems must be designed, tested, procured, deployed, and used in lawful and accountable ways so commanders and allies can trust the outputs. (defense.gov) Congress still has to decide what survives the budget process. But the direction in the request is already visible: more money for software that plugs into real missions, more infrastructure to test it, and less patience for AI demos that never leave the prototype stage. (theguardian.com)