Scale AI deploying first LLM on a government classified network

On May 10, San Francisco, CA-based Scale AI, the data infrastructure for AI, announced it is deploying the first large language model (LLM) on a classified network for the XVIII Airborne Corps – unlocking AI-powered decision-making for defense.

The partnership also marks the unveiling of Scale Donovan, the company’s federal AI platform to help operators, analysts, and decision-makers understand, plan, and act in minutes instead of weeks. Additionally, Donovan utilizes Reinforcement Learning From Human Feedback (RLHF) to continuously fine-tune the technology and better support mission objectives over time.

A key challenge facing military planners is the sheer volume of information available to support decision-making, most of which are in a variety of different formats. Now, decision-makers do not need to increase the size of their staff to manually analyze data. Using Donovan, they can make faster and better-informed decisions in rapidly evolving situations. The platform will be ingesting over 100,000 pages of live data for the XVIII Airborne Corps, such as orders, situation reports, and intelligence reports, to help military staff understand and organize the ever-growing volume of data with no training or coding experience – in minutes.

“To achieve decision advantage over adversaries in competition and conflict we must be able to condition, transform and aggregate data at the speed of war. LLMs are critical to our Corps vision of data centric warfare,” said COL. Joe O’Callaghan, the Fire Support Coordinator of the XVIII Airborne Corps. “We are thrilled to partner with the CDAO and Scale AI, in our data centric approach and operationalize this capability in our Scarlet Dragon series of exercises as we continue to refine our ability to out decide, out-think and out-perform any adversary.”

XVIII Airborne Corps will deploy Donovan for critical use cases:

  • Live context of friendly and enemy force operations by ingesting live data, such as orders, situation reports, and intelligence reports. This can help commanders better use the available staff, identify novel solutions more easily, and evaluate rapidly changing situations.
  • Context for intelligence-operations fusion by ingesting reporting, unit observations, and reducing the operational planning cycle time.
  • Decisional advantage with crisis planning support by ingesting open source data for potential conflict areas, which will identify insights to support intelligence preparation of the battlespace.

Donovan will also support the Joint All-Domain Command and Control (JADC2) to accelerate planning and enable the DoD to facilitate cross-combatant command collaboration. For example, the air tasking order cycle currently takes three days, and Donovan can shorten the timeframe of that cycle to hours.

Source: Scale AI

Stay in the know with breaking news from across the IC and IC contracting landscape by becoming a paid subscriber to IC News. Your support makes our work possible.