Scale AI graduates from labeling to targeting

When the Israel-Gaza war opened, one AI system produced more than 37,000 targets in the first weeks. That number — documented by The Guardian — is the closest thing to a ground truth the public has on what AI-assisted targeting actually looks like at scale. It is also the context that makes Tuesday's announcement from BAE Systems and Scale AI worth reading twice.

The two companies said they are combining forces to embed agentic AI into defense missions and platforms. Scale AI, a data-labeling startup that grew into one of the Pentagon's most consequential AI vendors, brings its Thunderforge program — which the Defense Innovation Unit has called the department's flagship effort for AI agents in military operations and planning. BAE Systems brings its Aided Target Recognition technology, which translates sensing into real-time coordinated effects across distributed forces. The partnership's pitch: close the gap between "the sensor sees something" and "the right unit acts on it," faster and at scale.

"The system will be deployed initially with U.S. Indo-Pacific Command and U.S. European Command," according to the contract announcement posted by the Defense Innovation Unit. Scale AI said it has spent the past two years doing prototype work with the DoD's Chief Digital and Artificial Intelligence Office before the September 2025 contract award.

Thunderforge is not a single system. It is an architecture — one that will embed Scale AI's agentic applications alongside Anduril's Lattice software platform and large language models from Microsoft, according to the DIU announcement. The program began with a $41 million initial award and carries a $100 million ceiling over five years. A separate $99.5 million Army research and development services contract, expected to run through 2030, was awarded to Scale AI. Peder Jungck, chief innovation and strategy officer for BAE Systems' Intelligence & Security sector, framed the AiTR integration as the sensory layer that makes the agentic stack actionable: "translates sensing into real time, coordinated effects across distributed forces."

That is the marketing language. The interesting question — the one Sky flagged on the board and that nobody in the press release answers — is what "agentic" means when the output is a target list rather than a travel itinerary.

Scale AI was co-founded in 2016 by Alexandr Wang and Lucy Guo, with Wang then 19 years old, to provide the data labeling infrastructure that powers machine learning systems. The company's original business was fundamentally human labor at scale — armies of annotators drawing boxes around objects in images and video. That proximity to AI training data gave Scale a view into where the DoD's AI ambitions were going before most defense contractors were paying attention. Zane Teeters, Scale AI's head of public sector go-to-market, is the other named contact on Tuesday's press release from BAE Systems and Scale AI.

The partnership with BAE represents Scale AI's most explicit pivot from "data labels" to "decisions." Aided Target Recognition is not new — BAE has worked on sensor-to-shooter automation for years. What changes with an agentic layer, as the press release describes it, is the tempo and distribution of the loop: not one sensor feeding one operator, but a mesh of sensing nodes running AI inference and coordinating effects across multiple platforms simultaneously. That is what "distributed forces" means in this context — and it is the design principle behind Thunderforge's architecture.

The Iran campaign offers a partial preview. Palantir's AI systems, which draw in part on large language model technology, were used to identify targets during the Iran campaign, according to multiple sources. One system alone produced more than 37,000 targets in the opening weeks of the conflict. Whether that output was a recommendation presented to a human analyst or something closer to an automated kill-list is the question the press release carefully avoids — and the question that determines whether "agentic AI in defense" is a coordination tool or an autonomous weapons accelerant.

This is the line the article is drawing: Scale AI and BAE are building the infrastructure for a world where the targeting cycle runs at AI tempo. The architecture is real, the contracts are funded, and the deployment locations are named. The question the DoD has not resolved — and that neither company answers in the announcement — is what a human being does in a loop that fast.

The Executive Order signed by President Donald Trump on September 5, 2025 restored the "Department of War" as an authorized secondary title for the DoD. Whether that institutional renaming signals a broader appetite for accelerated autonomous weapons development is a question beyond Tuesday's announcement. The partnership between BAE Systems and Scale AI does not need a policy shift to be significant. It needs a target.