Scale AI says 'tanks a lot' to Pentagon for data-classifying deal

First up: $41M to use human annotators to label all that unstructured military data. What could go wrong?

Data curation firm Scale AI has partnered with the Pentagon to deploy its AI on Top Secret networks - a move its interim CEO says is necessary if the US wants AI to be useful for national security.

The new Scale/Department of Defense deal is beginning with a $41 million award to Scale for advanced data labeling, the company told us, and has a ceiling of $100 million over five years. Along with providing labeling services, Scale is licensing three different applications to the DoD. First, there's its GenAI Platform - an end-to-end testbed for building, testing, and deploying defense and intel-specific AI models. Then there's its Donovan decision-making platform - a generative AI able to sift through unstructured data to suggest operational decisions "at mission speed." 

And finally, the DoD is getting access to Scale Data Engine, which it describes as a machine vision AI that can "turn … raw, sensitive data into the high-quality, AI-ready fuel needed to build reliable and effective models," the company said. 

"The promise of AI for national security can only be realized if it operates where the mission happens and on the most sensitive data," Scale AI's Jason Droege said in a press release provided to The Register. "This agreement bridges the critical gap between commercial innovation and the classified environment."

Scale has been working with the DoD for some time, as we reported early last year. The company said that it has spent the past two years doing prototype work with the DoD's Chief Digital and Artificial Intelligence Office, which led to today's contract. 

The Scale/DoD contract is just the latest in a series of agreements between AI firms and the US government, but Scale took pains to point out how different its deal with the so-called Department of War is from other government AI compacts announced in the past few months. 

For those that don't recall, the US General Services Administration has been on a buying spree since last month, inking deals for government access to AI models from OpenAI, Anthropic, Google, Microsoft, ServiceNow, and even Box. Those who haven't scored a deal yet - like Perplexity - have bent over backwards to try to get involved, too.

All of those deals have been short-term and based on deeply discounted terms for government agencies. Several companies, like OpenAI, have offered a year of access for $1 only to be undercut by Google at $0.47 and Microsoft, which was willing to give Copilot to certain agencies for free. 

Scale's DoD deal is similar in that it establishes department-level terms for the Pentagon to acquire and deploy Scale's platform (GSA's OneGov agreements likewise set standardized terms for federal agencies). However, the similarities end there, or so the company claims.

Unlike those other AI companies, Scale said that its deal offers an end-to-end AI platform instead of just a specific model. While Scale admits models are important tools, it argues that its contract with the DoD will give it a place to test and validate models on the department's data. Scale also posited that its agreement with the DoD to offer AI on Top Secret networks is its biggest differentiator.

It's not clear whether the GSA's broader, whole-of-government AI deals include more than just access to AI models. Information on those contracts has been limited, and the GSA has declined to share information when asked. 

Scale made headlines earlier this year not for the quality of its products or its work with the federal government, but for an entirely different reason: It was sued in a California court by contractors alleging exposure to disturbing content while training models. That lawsuit involved independent contractors working for Scale and AI-labeling contractor Smart Ecosystem in California. The $41 million initial award from the DoD will include getting access to "hundreds of specialists" at Scale's St Louis, Missouri AI Center, where the company has traditionally labeled geospatial data. 

The California case is ongoing; Scale is currently trying to push the matter into arbitration and a hearing is scheduled for next month. The AI company didn't respond to questions for this story. ®