News

The release ‘into the wild’ of PrimAITE – a Primary-level AI Training Environment created as part of Dstl’s multi-year Autonomous Resilient Cyber Defence (ARCD) programme – will enhance inter-governmental and agency collaboration, and enable a broader community of developers to contribute to and expedite the simulation’s evolution.

Already an in-demand capability across ARCD projects, the innovative tool – which was delivered through the Serapis SSE framework by a rainbow team led by QinetiQ – provides an effective means of training and evaluating AI and machine learning in a cyber-defensive role, with its customisable environment able to model platforms and system context by representing connections, IP addresses, ports, traffic loading, operating systems and services. Importantly, PrimAITE also operates at machine-speed to allow for fast training cycles.

“The decision to make PrimAITE open source will act as a force multiplier, helping to both advance other related research areas and fine tune the functionality and effectiveness of the software itself,” said James Short, a principal systems engineer at QinetiQ Training & Simulation and technical lead for the task. “The more the simulation can be honed, the better the AI agents it trains will become.

“Ultimately, this will assist in ensuring that military platforms can automatically identify and defend themselves against cyber-attacks, reducing the need for human intervention and accelerating response times.

“By limiting the risk of systems being compromised, ARCD will in turn protect those Service personnel who use them.”

Craig Murphy, Dstl’s technical lead for ARCD, said: “PrimAITE is an invaluable tool in our pursuit for autonomous cyber defence, allowing AI and machine learning experts to apply their skills and expertise to our problem domain. It’s now in use by more than 20 projects including other programmes in Dstl and with our international partners and we continue to develop it.”

Version 2.0.0 of PrimAITE is currently available on GitHub – a platform and cloud-based service for software development, with work underway to produce version 3.0.0, which will include enhanced simulation functionality and interoperability with agents.

To date, the QinetiQ Training & Simulation-designed – and Dstl/Ministry of Defence-owned –capability has been supported by specialists from Methods Analytics, Envitia and QinetiQ.