The Air Force's MAD.AI

Hey Kaitlyn, Joseph here. I heard the military liked AI. Now they're getting an AI for their AI. This piece is about a platform that aims to continuously improve the Air Force's existing systems in response to new environments. The full story and the documents follow below.

The Air Force is researching an AI tool that would continuously improve the AI systems on existing autonomous devices, such as drones and UAVs, helping them to respond to new environments, according to internal Air Force documents obtained by 404 Media.

The documents show the U.S. military’s continued adoption of AI, not just with generative AI tools based on OpenAI, but also tools that are marketed as improving internal AI systems even further.

This segment is a paid ad. If you’re interested in advertising, let's talk.

Generative AI companies are pushing the boundaries of creating realistic-looking video content from simple text prompts. But these innovations are also catching the attention of malicious forces—child predators, terrorist organizations, and hate groups are actively exploiting these new tools to further their dangerous agendas.

In ActiveFence’s latest report, we provide in-depth analysis of how these threat actors are weaponizing AI-generated videos. We reveal how child predators are bypassing safety measures to create video-based child abuse material. We examine how terrorists are incorporating AI video into their propaganda. Our report also explores strategies used by hate groups to produce videos that spread extremist ideology and radicalize viewers.

By understanding these tactics and technologies, companies and safety teams can better prepare to counter the misuse of generative AI—and protect their platforms from exploitation.

Read the report now.

“A Platform for Mission Adaptive AI (MAD.AI) for Networks of Autonomous Systems in Contested Environments,” the topic of the proposal, included in the documents, reads. 404 Media obtained the documents through a Freedom of Information Act (FOIA) request with the Air Force.

The company contracting with the Air Force on the effort is Qylur Intelligent Systems and cost $1.2 million, according to U.S. government procurement records. The documents say the contract is to provide a “modified and expanded” version of Qylur’s “Social Network of Intelligent Machines,” or SNIM AI. SNIM AI “is a software platform and enabling technology for maintaining the ongoing accuracy of autonomous systems AI,” according to Qylur’s website.

A screenshot of the documents. Image: 404 Media.

The purpose is to speed up the adaptive learning process and maximize the accuracy of AI models and their decisions, the website says. It adds that devices and machines can “evolve and adapt, by deploying ongoing Machine Learning across sites.” In short, it’s a tool to go along with a customer’s—in this case the Air Force’s—existing AI systems to improve them. In a press release from the time of the award, Qylur said the company “is exploring the adaptation of its SNIM® AI technology and how it can help the U.S. Air Force expedite the fielding of AI-based Collaborative Autonomous Systems, such as drones, UAVs, and ground systems.”

Qylur did not respond to a request for comment. But in a previous interview published on Forbes, Qylur CEO Dr. Lisa Dolev said the tool could help when, say, an object recognition system sees snow for the first time and everything looks different. SNIM AI could then take data from all of the connected machines, and use that to retrain the Air Force’s AI in response to those new conditions.

A screenshot of the documents. Image: 404 Media.

The newly obtained documents say that the Air Force needs SNIM AI to be modified in a number of different ways. One is to “expand data lake for extensive communications data.” The Air Force told 404 Media in an email that this is necessary “due to the amount of data generated from the countless permutations of communication environmental parameters for radio frequency data.  This research is not aimed at developing an adversarial capability.”

When asked if this system will be used in any combat situations, the Air Force told 404 Media that “any decisions regarding deployment have yet to be made.”