In the opening 24 hours of the US-Israel operation code-named “Epic Fury,” more than 1,000 targets in Iran were struck. US Central Command chief Gen. Brad Cooper said the broader campaign expanded to over 2,000 strike targets, with a scale nearly twice that of the 2003 Iraq war. The key difference, according to the source material, was not only size but tempo.
At the center of that tempo was not a single weapon but an AI-assisted workflow that compressed intelligence processing, target triage, and prioritization. The report says the military used Maven Smart System, developed by Palantir Technologies, with Anthropic’s Claude integrated as one of the core language models to interpret and organize large volumes of battlefield data.
More than 150 intelligence feeds pushed into one pipeline
The system reportedly ingested over 150 data sources, including satellite imagery, drone video, signals intelligence, human intelligence, and historical strike records. AI handled early-stage filtering and semantic interpretation, then assigned target priority levels before passing the results to human analysts.
Capt. Timothy Hawkins, a spokesperson for US Central Command, said AI could conduct initial screening of incoming data so analysts could focus on higher-level analysis and verification. In that setup, AI was not described as making the final strike call. It shortened the path between raw data intake and the point where humans could make an authorization decision.
From human timing to machine timing
The article contrasts this with the decision chain used during the Iraq war in 2003, when a target often took dozens of hours to move from intelligence confirmation to strike approval. Satellite images had to be reviewed frame by frame by people, and reports moved across departments for translation, briefing, and confirmation. Maven and Claude were described as compressing the most time-consuming part of that chain into machine-speed processing.
Using the figures cited in the report, 1,000 targets in a day works out to about 41 per hour. That number alone does not explain the shift. Each target still required intelligence collation, location verification, and threat assessment, which is why the source frames AI as changing the speed of war rather than simply adding automation.
Claude was banned by the federal government and still remained in the chain
The report also highlights a political and operational contradiction. On February 27, 2026, the Trump administration announced a ban on Anthropic, ordering federal agencies to stop using Claude over what it called supply-chain and national security risks. At nearly the same time, Claude was still being used in the Epic Fury workflow for strike target analysis.
The Pentagon later said there was a six-month transition period. The article argues that once an AI model is deeply embedded in an active combat workflow, replacing it in the middle of operations is not a simple technical swap. That gap between policy directives and operational dependence is one of the clearest points raised in the piece.
Defense contracts, platform integration, and model replacement
Maven Smart System traces back to the Pentagon’s Project Maven in 2017 and was later carried forward by Palantir. The report says Claude’s presence on Pentagon classified networks followed a 2024 contract between Anthropic and the Department of Defense, with a two-year term and a ceiling of roughly $200 million.
The same source says that if Claude were removed, Maven could still connect to OpenAI models, and that OpenAI moved quickly to announce it would take over the Pentagon’s AI systems after the Anthropic ban. Based on the material provided, the constant is not a single vendor. It is the operating logic: AI compresses the intelligence workload, while human officers retain the final authorization step.