This week’s defence AI stories reveal a sector in rapid, contested transition: Europe is racing to field autonomous drone swarms while the US Air Force validates edge AI chips for live combat missions, and a landmark Senate bill attempts to draw ethical red lines that the Pentagon is simultaneously pushing against. The Anthropic-Pentagon legal standoff — now playing out in federal courts — is crystallising the defining question of the era: who controls the kill switch on military AI?
Top story: Europe bets €90m on 50,000 AI-enabled drone swarm as Germany, UK, France and Norway signal intent to scale autonomous warfare capacity.
Europe’s Drone Arms Race: Germany Orders 50,000 AI-Enabled Swarm Drones
CNBC · Strategy
Defence software company Auterion and Ukrainian drone maker Skyfall announced a 90-million-euro order for 50,000 AI-enabled drones from a European NATO member, confirmed to be Germany. The technology is already attracting interest from the armed forces of the UK, Norway, and France. The deal signals that battlefield lessons from Ukraine have accelerated European willingness to procure AI-autonomous drone systems at industrial scale, with Helsing and other software-first defence firms emerging as the new power brokers.
https://www.cnbc.com/2026/07/15/drones-defense-europe-autonomous-weapons.html
US Air Force Flies Japanese Edge AI Chip in Live Combat Exercise
Interesting Engineering · Tools
The US Air Force, backed by the Defense Innovation Unit, has validated the EdgeCortix SAKURA-II AI accelerator chip in a live large-force exercise — the first demonstration of a Japanese-designed edge AI processor running military inference workloads in flight. The chip delivered high-performance AI processing with minimal power consumption, without relying on cloud connectivity, a critical requirement for contested airspace. The milestone points to a growing allied-nation semiconductor supply chain for battlefield AI, reducing dependence on US-only hardware.
https://interestingengineering.com/military/edgecortix-sakura-ii-us-air-force-edge-computing
Senate Democrats Table ‘Responsible AI Defence Act’ to Govern Pentagon’s AI
AI Risk Today · Regulation
Senate Democrats introduced the Responsible Artificial Intelligence Defense Act (S. 4707), which would require the Pentagon to classify military AI systems by risk tier and subject higher-risk systems to legal, cybersecurity, and privacy reviews before deployment. A central provision would prohibit AI from making nuclear launch decisions, and would restrict autonomous systems from tracking or targeting individuals inside the United States without legal authorisation. If enacted, it would create the first formal statutory framework for military AI governance, requiring annual reports to Congress through 2037.
https://www.airisktoday.com/responsible-ai-defense-act-2026/
Pentagon’s GenAI.mil Hits 1.7M Users as ChatGPT Joins Classified Network
Defense One · Generative AI
The Pentagon’s GenAI.mil platform has reached almost 1.7 million military users, with OpenAI’s ChatGPT now cleared to handle controlled, unclassified information through the platform from July 2026. Defence officials say one of the primary use cases is aggregating data for warfighters to accelerate decision-making, with agentic tools operating under ‘very tight guardrails’ to assist analysis that would otherwise require two or three human analysts working across disparate systems. The scale of adoption marks a significant milestone in routine military reliance on frontier commercial AI models.
AI-Assisted Cyber Attack Achieves Full Cloud Compromise in 72 Hours
Enterprise Times (UK) · Risk
UK cybersecurity firm Sygnia released initial findings from an investigation into an attack where threat actors used AI-assisted tooling to move from initial AWS access to full environmental compromise in roughly 72 hours. The AI tools enabled attackers to chain multiple techniques and access keys with minimal human effort, dramatically compressing the traditional intrusion timeline. The case is being cited as a landmark illustration of agentic automation risk — not just for enterprises, but for defence supply chains where cloud environments host sensitive government data.