{"id":652942,"date":"2026-05-06T08:45:14","date_gmt":"2026-05-06T08:45:14","guid":{"rendered":"https:\/\/www.newsbeep.com\/au\/652942\/"},"modified":"2026-05-06T08:45:14","modified_gmt":"2026-05-06T08:45:14","slug":"australian-defence-ai-policy-risks-writing-modern-ew-out-of-the-force","status":"publish","type":"post","link":"https:\/\/www.newsbeep.com\/au\/652942\/","title":{"rendered":"Australian Defence AI policy risks writing modern EW out of the force"},"content":{"rendered":"

<\/p>\n

In 2025, Ukrainian-led opposing forces defeated NATO formations in two major exercises. At exercise Hedgehog in Estonia, about 10 drone operators with cheap first-person-view drones rendered two NATO battalions combat-ineffective in half a day. At Dynamic Messenger off Portugal, a Ukrainian-led red team beat NATO naval forces in all five scenarios, sinking a frigate undetected \u2013 with low-cost systems that cost a fraction of the platforms they destroyed.<\/p>\n

NATO forces couldn\u2019t detect or counter them. The electromagnetic warfare (EW) capabilities that could have done so \u2013 software-defined, AI-enabled systems able to identify and disrupt drone control links across a wide area \u2013 were either not present or inadequate.<\/p>\n

These are known as cognitive EW systems. The technology exists, and indeed my company works on it. A failure to field cognitive EW is an adaptation problem, and Australia risks enshrining the problem into policy.<\/p>\n

Cognitive EW systems don\u2019t operate from pre-loaded threat libraries that match known signals with programmed responses. Instead, they observe the electromagnetic environment, classify what they detect, select a response \u2013 and improve that process with every encounter. Their models for judging threats evolve by the hour rather than, as is common for traditional systems, updating quarterly. The capability that matters isn\u2019t classification of targets, which is well-understood; it is selecting a response to signals the system has never seen before.<\/p>\n

This is a structural issue. Cognitive EW systems can\u2019t be specified against the fixed requirements routinely set in acquisition programs, because their behaviour changes with every deployment. Acceptance milestones assume a system\u2019s behaviour can be characterised and locked in. Cognitive EW systems \u2013 which observe, orient, learn, decide and act \u2013 are designed to defy that assumption.<\/p>\n

The US Army appears to have recognised this. In February, it abandoned detailed specifications for EW and instead issued a broad \u2018characteristic of need\u2019 for electromagnetic spectrum operations, inviting industry to propose capabilities against problem sets rather than requirements. Its acquisition executive for command and control told<\/a> Breaking Defense that being \u2018hyper specific about what it needs to do\u2019 was failing to recognise the opportunity software-defined architectures offer.<\/p>\n

Australia\u2019s framework is moving the other way. The Australian policy\u2019s risk-based, proportionate framing is the right foundation, and its explicit inclusion of frontier models is more forward-looking than many allied equivalents. The difficulty is in how obligations for traceability and accountability will operate in practice for systems designed to learn in the field. Defence\u2019s Responsible AI policy<\/a> requires that \u2018the function and relationship between inputs and outputs of AI technologies must be traceable\u2019 and makes individual Defence officials personally accountable for \u2018the actions, outcomes and decisions of AI-enabled functions\u2019.<\/p>\n

For a cognitive EW system that learns in the field, this standard is unachievable in practice. Even the developers of today\u2019s leading AI models, with their eye-watering research budgets, cannot trace inputs to outputs through the billions or trillions of parameters the models contain. The accountability mechanism will rationally discourage officials from fielding systems whose behaviour they can\u2019t fully predict.<\/p>\n

This standard is appropriate for AI in a kinetic kill chain. It isn\u2019t fit for purpose for unarmed autonomous systems, including those providing non-lethal electromagnetic attack. A cognitive EW jammer that disrupts a drone\u2019s control link produces a reversible electromagnetic effect, yet Australia is subjecting it to the same acquisition rules as an autonomous kinetic weapon.<\/p>\n

Non-lethal doesn\u2019t mean zero-consequence. A jammer can disrupt civilian communications. But the risk profile is fundamentally different from a system that selects human targets, and governance should reflect that.<\/p>\n

Innovation units within the ADF and government departments fund research into these technologies. But they can\u2019t fix the acquisition and sustainment frameworks capabilities must pass through to reach the force, nor can they selectively interpret the AI policy that governs them once they arrive.<\/p>\n

Australia should consider whether its approach is fit for purpose when the forces it interoperates with are abandoning it. As the US Army moves to continuous delivery and broad problem statements while Australia is tightening fixed requirements and traceability expectations, the two forces will struggle to integrate their electromagnetic capabilities. It\u2019s an interoperability risk.<\/p>\n

The policy already requires Defence personnel to exercise \u2018informed judgement and care\u2019 and anchors accountability in \u2018reasonably anticipated impacts\u2019 \u2013 foreseeability standards familiar from other domains.<\/p>\n

Extending this logic, accountability for cognitive EW systems should rest on the rigour of testing, verification and validation, and on how human supervision is structured, not on predicting behaviour that by design cannot be predicted. Defence already navigates this balance for soldiers, whose accountability attaches to the reasonableness of their conduct rather than to outcomes they could not reasonably have foreseen. AI should sit in the same frame.<\/p>\n

Working out what that looks like in practice is the task ahead \u2013 one the Defence Artificial Intelligence Centre is positioned to lead, and one that industry is ready to support.<\/p>\n","protected":false},"excerpt":{"rendered":"In 2025, Ukrainian-led opposing forces defeated NATO formations in two major exercises. At exercise Hedgehog in Estonia, about…\n","protected":false},"author":2,"featured_media":652943,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[20],"tags":[256,254,255,64,63,105],"class_list":["post-652942","post","type-post","status-publish","format-standard","has-post-thumbnail","category-artificial-intelligence","tag-ai","tag-artificial-intelligence","tag-artificialintelligence","tag-au","tag-australia","tag-technology"],"_links":{"self":[{"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/posts\/652942","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/comments?post=652942"}],"version-history":[{"count":0,"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/posts\/652942\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/media\/652943"}],"wp:attachment":[{"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/media?parent=652942"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/categories?post=652942"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/tags?post=652942"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}