This week’s defence AI landscape is defined by a widening gap between rapid deployment and governance — the US Air Force is flying AI chips in live exercises while Congress presses Pentagon tech contractors to disclose what their AI is actually authorised to do. Europe is moving in parallel, with the EU’s cybersecurity AI plan now entering enforcement and Brussels candidly admitting its dependency on US AI models for its own defence.
Top story: The US Air Force flew a Japanese AI edge-computing chip inside a live large-force exercise, marking the first validated airborne deployment of commercial AI inference hardware in an operational military mission system.
US Air Force Flies Japanese AI Chip in Live Combat Exercise
Interesting Engineering · Tools
The US Air Force validated EdgeCortix’s SAKURA-II edge AI chip during a large-force exercise, integrating it into a prototype gateway designed to provide resilient multi-domain battlefield connectivity across joint and coalition forces. The chip delivered high-performance, low-power AI inference in flight — a significant step toward truly autonomous, on-aircraft AI decision-making. For defence practitioners, it signals that off-the-shelf commercial AI silicon, not bespoke military hardware, may define the next generation of airborne intelligence.
https://interestingengineering.com/military/edgecortix-sakura-ii-us-air-force-edge-computing
Senate Presses Pentagon and Big Tech to Reveal AI Contract Terms
Federal News Network · Regulation
A senior US senator demanded that the Department of Defense and major AI vendors disclose the terms of their classified AI contracts, focusing on whether provisions against mass surveillance and autonomous lethal targeting are actually enforceable. Critics warned that current contract language is so broad the government can ‘drive a truck through’ the safeguards. The scrutiny follows the Senate Armed Services Committee adding language restricting military AI use to the 2027 defence policy bill, putting AI governance at the heart of Pentagon procurement.
EU AI Cybersecurity Plan Hits Enforcement Date With Key Gaps Exposed
Euronews · Regulation
The EU’s AI Act provisions underpinning its cybersecurity action plan moved into enforcement on 2 August 2026, but Euronews revealed a critical vulnerability: Brussels is largely dependent on negotiated access to US AI models for its own cyber-defence capabilities, with little sovereign European capacity to substitute them. The plan builds on the AI Act, NIS2, the Cyber Resilience Act, and the Cyber Solidarity Act, but critics note the Commission has offered mostly recommendations rather than binding obligations. For European security and compliance teams, the enforcement clock is now running while the tools to meet it remain largely American.
US Air Force Unveils AI-First Strategy for Battlefield Decision Advantage
GovConWire · Strategy
The Department of the Air Force published its AI-First Strategy, making automated multi-level security a centrepiece of how warfighters will access classified information at the tactical edge. The service is simultaneously building a fleet of Collaborative Combat Aircraft — AI-piloted drones designed to operate alongside manned fighters for air-to-air combat, targeting, and electronic warfare. Data silos across classification levels remain the primary technical barrier, but the strategy signals that AI-driven autonomy in the cockpit is now an institutional priority, not an experiment.
https://www.govconwire.com/articles/air-force-ai-multi-level-security-edge
EU Cyber Plan Exposes UK-EU AI Defence Rulebook Split Post-Brexit
Bratby Law / techUK · Risk
An analysis published by Bratby Law for techUK this week highlighted that the EU’s cybersecurity and AI action plan creates a two-rulebook problem for any business operating critical systems in both the UK and EU — forcing compliance with overlapping but distinct AI and cyber frameworks simultaneously. The EU plan establishes a secure ENISA testing platform for critical infrastructure operators to validate AI before deployment, expected by Q4 2026, while the UK pursues the same goals without a single horizontal AI law. For UK defence contractors and dual-jurisdiction operators, the divergence is now a live compliance risk, not a future planning assumption.