Project Maven emerges as key Pentagon AI programme in US strikes on Iran

WASHINGTON: Project Maven, the Pentagon’s main artificial intelligence programme, has moved to the forefront of the United States’ strikes on Iran and is being seen as part of a major shift in the conduct of modern warfare.

The programme began in 2017 as a limited effort aimed at helping military analysts process the huge volume of drone footage coming from conflict areas. At the time, analysts were reviewing imagery frame by frame to spot objects of interest that could appear only briefly before disappearing. The system was designed to help identify such targets more quickly.

Over the past eight years, Project Maven has developed into a much broader AI-supported system for targeting and battlefield management. This has significantly sped up what is known as the kill chain, the sequence that runs from initial detection of a target to its destruction.

Aalok Mehta, director of the CSIS Wadhwani AI Center, described Maven as “essentially an overlay” that combines sensor information, intelligence on enemy troops, satellite imagery and data on troop deployment. In practical terms, the system can quickly examine satellite feeds to detect troop movements or identify targets while also “taking a snapshot of the operational theater” to help determine the most effective option for striking a specific target.

In a recent online demonstration, a Pentagon official said Maven “magically” converts an observed threat into a targeting workflow, assesses available assets and presents commanders with possible options.

The arrival of ChatGPT marked another important step in the programme’s development, expanding the technology’s use to a much wider group of users who can interact with Maven through natural language. At present, that capability is provided by Anthropic’s Claude. However, the arrangement is ending after the Pentagon objected to the AI company’s requirement that its model not be used for fully automated strikes or for tracking US citizens.

Ethical concerns also shaped the programme’s early history. Google was Maven’s original AI contractor, but in 2018 more than 3,000 employees signed an open letter opposing the company’s role, saying the contract crossed an ethical boundary. Several engineers also resigned.

Google did not renew the contract when it expired and later issued AI principles that explicitly excluded involvement in weapons systems. The episode highlighted divisions in Silicon Valley between engineers who considered autonomous targeting unacceptable and defence officials who regarded it as necessary.

More recently, Google removed its AI policy restrictions and stated that it is moving further into national security work. The Pentagon has said Google, xAI and OpenAI are among those being considered to replace Claude in Maven.

In 2024, Palantir entered the area left by Google. The company, founded in part with CIA seed funding and built around government intelligence work from the outset, has reportedly become Maven’s main technology contractor. Its AI now serves as the programme’s operational backbone.