IN Brief:
- The US Army hosted AI Table Top Exercise 2.0 with 14 technology companies.
- The exercise explored AI-enabled cyber defence in an Indo-Pacific conflict scenario.
- The next phase is expected to focus on rapid pilots, testing labs, autonomous agents, and secure deployment.
The US Army is preparing a fast follow-up to an AI cyber defence wargame involving executives from 14 technology and cybersecurity companies.
AI Table Top Exercise 2.0 used a 2027 Indo-Pacific crisis scenario in which an adversary launched repeated AI-enabled cyber attacks against Army networks. The exercise examined how automated tools, frontier models, and AI agents could support defensive operations when human response times are too slow.
Participants included Amazon Web Services, Google, Microsoft, OpenAI, CrowdStrike, Palo Alto Networks, and other technology companies, alongside Army and Department of War leadership. The focus was defensive: network protection, automated response, deception techniques, and methods for countering hostile AI agents.
The next phase is expected to involve rapid pilots inside Army Cyber Command testing environments. Cycles of 30 to 90 days would allow the service to assess commercial tools more quickly than traditional acquisition routes.
Secure deployment in compressed cycles
AI cyber tools cannot enter military networks as standard commercial products. They must be integrated into secure environments, tested against realistic attack behaviour, and governed by rules that define where automation stops and human control begins.
That places pressure on model safety, explainability, audit trails, deployment architecture, and operator interfaces. Defensive cyber teams need tools that can assist at machine speed while leaving commanders with enough visibility to trust the system’s actions.
Legacy integration will also shape adoption. New AI agents must work alongside existing sensors, logging platforms, identity systems, classified networks, and cyber mission workflows. A powerful model that cannot be securely deployed or monitored has limited operational value.
The Army’s approach could create a faster procurement route for cyber suppliers, but the production standard remains demanding. Companies moving beyond pilot projects will need secure software pipelines, realistic testing evidence, compliance controls, and support models that can survive operational use rather than laboratory demonstrations.
