Independent intelligence on AI
SectorFrame topic
Back to Defence & SecurityAI-supported situational awareness, resilient command systems and cyber operations.
The lead story
WIRED reports that a contractor connected a personal coding agent to a sensitive defence system. Accenture disputes the allegations. The account raises questions about delegated credentials, external services and responsibility for incident reporting.
Project Gatekeeper lets companies run their code through the US Army's security checks in minutes rather than waiting months for approval. Frontier AI models are to be added next, though the Army has not said whose.
Helsing has unveiled Contra, an AI-run air-defence system, and two container-launched interceptors it says will undercut existing missiles on price. The bigger ambition is the command layer that decides which interceptor fires.
A C4ADS study of Russian trade records finds the Rubin design bureau sent more than $80m of submarine-related shipments to Chinese military buyers between 2015 and 2023. The paper trail is thin on detail but long on duration, and it ran both ways.
The Japanese provider is seeking government intervention after a breach exposed trading credentials. Its response reveals how a cloud service can become a financial-security problem, while the role of AI remains unproven.
Ukraine's defence ministry says Starlink and Starshield are stable again above 400km/h after ten days of restrictions on fast-moving platforms. Nobody has said why the limits went on or how they came off, which is the part planners should note.
A 20 October acquisition memo gives military buyers mandatory templates for enforcing the White House's Unbiased AI Principles. Vendors must document how their models behave, and the obligation reaches back towards the labs that built them.
Agents set to attack OpenAI’s own training enclosure found an unauthenticated code-execution bug in Brotli, the library inside Node.js, Envoy and Python. The fix had sat in the source code since February, unreleased and unlabelled.
Moonshot says it stripped cyber data from Kimi K3.1 Pro before its weights ship on 4 November. Its own test shows a modest dent; Artificial Analysis ranks it the strongest openly available model for defensive security work.
The US Navy has awarded Applied Intuition an initial $25m to field its Intelligent Perception System on five Arleigh Burke-class destroyers. The sensor fusion is useful; the larger change is that warships start collecting data to retrain the models that run on them.