This week’s defence AI landscape is defined by a race between ambition and governance: the UK published its landmark Defence Investment Plan committing over £5 billion to autonomous systems and AI infrastructure, while Europe grapples with a structural gap in military AI oversight as the EU AI Act explicitly exempts defence applications. Globally, the convergence of commercial frontier AI with classified military networks — and mounting controversy over human-control requirements — is reshaping how democracies procure, deploy, and ethically constrain lethal AI systems.
Top story: The UK’s £298 billion Defence Investment Plan, published 30 June 2026, earmarks over £5 billion for drones and autonomous systems and £7.3 billion for a ‘digital targeting web’ — the most significant UK defence AI commitment in a generation.
UK’s £298bn Defence Plan Bets Big on AI and Autonomous Systems
House of Commons Library / Chatham House · Strategy
On 30 June 2026, the UK government published its long-awaited Defence Investment Plan, committing more than £5 billion to drones and autonomous systems across the armed forces over the next four years, alongside £7.3 billion for a ‘digital targeting web’ connecting sensors, AI decision-support, and strike assets. The plan also includes £2.5 billion for a new Defence Cyber and Electromagnetic Force and £115 million specifically to raise UK defences against AI-related threats. For defence practitioners, this signals that AI and autonomy have moved from aspiration to funded acquisition priority — with implications for procurement, industry partnerships, and readiness standards across the entire UK military.
https://commonslibrary.parliament.uk/research-briefings/cbp-10935/
Europe’s AI Act Leaves Military AI in a Dangerous Regulatory Void
CEPR VoxEU · Regulation
A CEPR analysis warns that Europe’s defence ministries and intelligence services run on AI infrastructure they cannot independently audit, a vulnerability sharpened by a June 2026 US export control directive ordering Anthropic to suspend access to its most powerful models for all non-US persons. The EU AI Act deliberately excludes military and national security AI, meaning Europe has built no instrument to govern its own military AI — and none to govern anyone else’s — leaving it as an observer rather than a party when Washington makes unilateral moves. The piece argues it will take ten to fifteen years for Europe to develop sufficient sovereign AI capability to negotiate as a peer, and proposes interim steps including formal political assessments and notification agreements with Washington.
https://cepr.org/voxeu/columns/europes-ungoverned-space-military-ai-and-autonomy-cannot-be-bought
New Zealand Joins Five Eyes’ AI Battlefield Command Network
The Defense Post · Strategy
New Zealand has become the latest Five Eyes member to join Project Arcadia, a US-led initiative to build a shared battlefield command network for AI-enabled military operations. The move deepens the alliance’s push toward interoperable, AI-assisted command-and-control across its core intelligence-sharing partners. For security professionals, this illustrates how allied nations are rapidly standardising AI decision-support infrastructure at the coalition level — raising important questions about shared governance, accountability, and the rules of engagement for AI-assisted targeting across multi-national forces.
Pentagon’s ‘Any Lawful Use’ AI Contracts Alarm European Allies
Security Affairs · Risk
The US Department of War has signed agreements with major tech firms — including OpenAI, Google, Microsoft, Amazon, and SpaceX — to deploy frontier AI on classified military networks under ‘lawful operational use’ clauses, explicitly requiring vendors to accept any Pentagon-sanctioned use, including autonomous weapons. More than 1.3 million DoD employees are already using the GenAI.mil platform. The removal of human-control requirements from US military AI doctrine, combined with the exclusion of Anthropic after it refused to drop safety guardrails around autonomous weapons, has drawn sharp concern from European partners who fear being locked out of oversight over AI systems affecting their own assets and networks.
LLMs Are Lowering the Barrier to CBRN Terrorist Attacks, Europol Warns
Stimson Center · Risk
A Stimson Center analysis, drawing on a Europol red-team exercise, finds that large language models are materially increasing the ability of non-experts to plan chemical, biological, radiological, and nuclear attacks by unlocking specialised datasets and reducing the time needed to develop credible attack scenarios. A comparative study of ten leading commercial LLMs found an 87-percentage-point gap in bio-security performance between the most and least secure models, underscoring that mitigation is technically achievable but inconsistently applied across the industry. The findings add urgency to calls for structured public-private partnerships between AI developers and the counter-terrorism community, and will inform how security agencies assess frontier model releases going forward.
https://www.stimson.org/2026/harnessing-ai-to-move-from-threat-to-threat-reduction/