Professionals working on laptops and computers

It is vital for NATO and its partner organizations to be able to rapidly analyze and use information that is relevant to security. We have worked with NATO’s Allied Command Transformation (ACT) to develop and implement its first AI-supported modular “knowledge agent” within the scope of a lighthouse project with the BI-SC NATO SECRET (NS) security classification level. The solution works without an Internet connection, significantly speeds up the involved processes and delivers highly precise results.

 

NATO’s first widely used AI tool: AI FELIX

ACT, based in Norfolk, Virginia in the United States, is one of NATO’s two strategic military commands. It is responsible for driving the alliance’s future development. Every day, hundreds of incoming documents need to be analyzed, compiled, classified, edited and used. AI FELIX, a digital “knowledge agent,” was created to significantly accelerate these processes while also greatly improving access to data and achieving information superiority. Within the scope of an internal project that began in 2019, ACT created the first prototype and minimum viable product (MVP). The result was a very successful example of ACT’s vision to introduce AI-based process automation.

In 2023, from that base, ACT and we then launched a joint effort to refine, scale and integrate the MVP into a fully fledged industrialized enterprise AI service. By combining ACT’s vision with lessons learned and our expertise in developing and rolling out enterprise applications on a large scale, we succeeded in rapidly implementing the first module – the Command Read Board (CRB) – across 27 locations by the fall of 2024. Our intimate familiarity with NATO’s processes was also crucial for the success of this project.

ACT needed two additional modules to be introduced by the end of 2024: Task Portfolio Management (TPM) and a customized AI-enabled digital assistant (FELIX).

AI FELIX at a glance

  • Provision of NATO’s first broadly used AI capability
  • NS security level: without Internet connection, completely implemented on-premises
  • Implemented so far at 27 locations of the NATO command structure
  • Scalable for up to 30,000 users
  • Processing time per document reduced from 5 – 10 minutes to 27 seconds
  • Use of state-of-the-art technologies such as machine learning and advanced analytics
Two professionals working in front of a computer screen
27 seconds
processing time per document reduced from 5–10 minutes
27 locations
of the NATO command structure have implemented the solution
Above view of hands typing on laptop surrounded by documents

Command Read Board (CRB): the central information processing platform

One of the biggest challenges faced while developing and implementing this AI-supported tool was creating the protected environment, separate from the Internet, in which the CRB executes.

After receiving the first prototypes from NATO, we trained, fine-tuned and developed them further on tens of thousands of documents that had been generated specifically for this purpose. The algorithms involved were also tweaked to optimize their functionality.

Today the CRB is able to analyze incoming documents, emails and other messages, generate summaries, classify correspondence according to its confidentiality level and extract metadata from it, identify and define tasks, prioritize information and much more.

CRB does not operate in isolation; instead, it is fully integrated in NATO’s suite of information management tools, which include electronic document management system (EDMS) and task trackers. It is also straightforward to adapt to other environments in NATO’s member countries, for example Germany’s DokMBw.

Coming next: TPM for task portfolio management and FELIX for providing customized information

The modules of AI FELIX are interlinked. CRB identifies and generates tasks based on incoming documents, among other things. TPM helps keep track of these tasks, quickly identifies potential risks and generates tailored reports on these activities. Specifically, it monitors and controls tasks and makes predictions with AI support.

The third module, FELIX, flexibly provides users with information that is aligned with their particular responsibilities, needs and interests. FELIX can refer a user to certain documents or training courses or else suggest that they get in touch with other users. It also includes a “light” version of CRB that calls their attention to specifically relevant documents, automatically summarizes content or generates metadata.

Two professionals working in front of computer screens
Two professionals analyzing data in front of computer screens

AI at NATO: a strategy with a future

NATO ACT is convinced that the challenges the alliance is facing can only be mastered via an ongoing process of innovation. Especially for digitalization, it therefore makes a great deal of sense to use an artificial intelligence such as AI FELIX that can be quickly implemented, also in closed networks.

AI FELIX began as a pilot effort and proof of concept but is now achieving maturity. While taking an agile approach, we will continue to steadily refine it in close coordination with users and develop and implement more new tools to facilitate everyday work and improve the quality of data.

Additionally, to achieve a scalable AI FELIX solution across 27 NATO locations, the solution is built on CGI AIOps Director, a platform for scalably deploying AI services and cloud technologies, also featuring secure air-gapped infrastructure.