When the consumer asks “What enemy navy unit is within the area?” the AIP Assistant guesses that it’s “seemingly an armor assault battalion primarily based on the sample of the tools.” This prompts the analyst to request a MQ-9 Reaper drone to survey the scene. They then ask the AIP Assistant to “generate three programs of motion to focus on this enemy tools,” and inside moments, the assistant suggests attacking the unit with both an “air asset,” a “long-range artillery,” or a “tactical group.” The consumer tells the assistant to ship these choices to a fictional commander, who finally chooses the tactical group.
The ultimate steps play out rapidly: The analyst asks the AIP Assistant to “analyze the battlefield,” then “generate a route” for troops to achieve the enemy, and eventually “assign jammers” to sabotage their communications tools. Inside seconds, the analyst provides the battle plan a ultimate overview and orders the troops to mobilize.
On this state of affairs, Claude can be the “voice” of the AIP Assistant, and the “reasoning” it makes use of to generate responses. Different AIP demos present customers interacting with giant language fashions in a lot the identical means. In a weblog published final week, for instance, Palantir detailed how NATO, a Maven Good Methods customer, might use an AIP Agent throughout the instrument.
In a single graphic, Palantir exhibits how a third-party protection contractor can choose from a number of of Palantir’s built-in AI fashions, together with totally different variations of OpenAI’s ChatGPT and Meta’s Llama. The consumer selects OpenAI’s GPT 4.1, however seemingly, this could possibly be the place a soldier would even have the choice to select Claude as a substitute.
An analyst then views a digital map displaying the places of troops and weapons. In a panel labeled “COA” (programs of motion), they click on a button that prompts a instrument powered by GPT-4.1 to generate 5 doable navy methods, together with one known as Assist-by-Fireplace-Then-Penetration-Shock-and-Destruction.
One other instance exhibits how the system might assist interpret satellite tv for pc imagery: The analyst selects three tanker truck detections on a map, hundreds them into the AIP Agent’s chat interface, and asks it to “interpret” the imagery and counsel choices for what to do subsequent.
Claude might also be utilized by the navy to create intelligence assessments which will inform strike planning later down the road. In June 2025, WIRED considered an indication given by Kunaal Sharma, a public sector lead at Anthropic, displaying how the enterprise model of Claude could possibly be used to generate “superior” experiences about an actual Ukrainian drone strike dubbed Operation Spider’s Web. Within the demo, Sharma defined, Claude was relying solely on publicly accessible data. However by partnering with Palantir, he mentioned, the federal authorities can even pull from inside datasets.
“That is usually one thing that I would sit for like 5 hours with a cup of espresso, and skim Google, and go into suppose tanks, and begin writing experiences and writing a quotation, and so forth.,” Sharma mentioned. “However I haven’t got that form of time.”
Within the demo, Sharma requested Claude to create an “interactive dashboard” with details about Operation Spider’s Internet, after which translate it into “object sorts” that could possibly be analyzed in Foundry, one in all Palantir’s off-the-shelf software program merchandise. He additionally requested Claude to put in writing an in depth evaluation of current developments in Russia’s border provinces, in addition to a 200-word synopsis of the operation’s “navy and political results.”
“Frankly, I have been studying all these issues for 20 years—I used to put in writing them, I was an educational myself,” Sharma mentioned, “That is really fairly good.”
