$35K Kamikaze Drones See First Combat as Pentagon Bans Anthropic and Arms Its Autonomous Wingmen

This was the week autonomous warfare stopped being theoretical. The U.S. military deployed low-cost kamikaze drones in combat for the first time, the Pentagon banned its primary AI provider and replaced it within hours, autonomous fighter prototypes flew with live weapons mounts, and Taiwan published a drone swarm defense doctrine. Every one of these stories points in the same direction: cheap, autonomous, edge-deployed systems are now the center of gravity in defense technology.

LUCAS Drones Draw First Blood

The biggest story of the week is the combat debut of the Low-cost Unmanned Combat Attack System (LUCAS) during Operation Epic Fury. CENTCOM confirmed that Task Force Scorpion Strike deployed LUCAS drones against Iranian Revolutionary Guard Corps command nodes, air defense sites, and military airfields—marking the first time the U.S. military has used one-way attack drones in combat.

The LUCAS is a reverse-engineered version of Iran’s Shahed-136, built by Arizona-based SpektreWorks at roughly $35,000 per unit. For context, a Tomahawk cruise missile costs approximately $2 million. The cost ratio is roughly 57:1.

But cost isn’t what makes LUCAS strategically significant. The drone features a Multi-domain Unmanned Systems Communications (MUSIC) mesh network that enables autonomous swarm coordination and network-centric strikes. Some units are equipped with Starlink terminals for dynamic target acquisition. The system is designed to be operated by non-specialized personnel, eliminating the extensive training pipeline that traditional weapons systems require.

The doctrinal implications are profound. CENTCOM didn’t just use drones—it stood up a dedicated task force (Scorpion Strike) for one-way attack operations, signaling this is a permanent capability, not a one-off experiment. The message: saturation with affordable autonomous systems is now a core U.S. combat strategy.

The Pentagon-Anthropic Breakup—and OpenAI’s Rapid Replacement

The AI governance story that had been building for weeks reached its crescendo. After Anthropic refused to remove safety guardrails around domestic mass surveillance and fully autonomous weapons from its contract terms, the Pentagon designated the company a “supply-chain risk to national security.” This designation means no military contractor or supplier can do business with Anthropic—a corporate death sentence in the defense market.

The timeline was surgical. The administration ordered all government agencies to stop using Anthropic’s technology. Thirteen minutes after a Pentagon-imposed deadline expired, the ban went into effect. Hours later, OpenAI announced it had reached an agreement to provide AI technology for classified DoD networks.

OpenAI’s deal reportedly includes the same two safety principles Anthropic fought for—prohibitions on domestic mass surveillance and human responsibility for the use of force. The difference: OpenAI negotiated them as voluntary commitments rather than contractual red lines. Whether that distinction holds under operational pressure remains to be seen.

The broader signal is unmistakable. The Pentagon wants AI providers who deliver capability without contractual friction. Companies that attach conditions to military use of their technology risk being replaced—quickly. This is reshaping how every AI company in the defense ecosystem evaluates its government strategy.

Armed CCA Prototypes Take Flight

While the world watched Iran, the Air Force quietly hit a major milestone in its Collaborative Combat Aircraft (CCA) program. Anduril’s YFQ-44A—now officially named “Fury”—completed captive carry tests with an inert AIM-120 AMRAAM missile, validating the airframe’s ability to carry weapons in flight.

The CCA program is accelerating on multiple fronts simultaneously. General Atomics’ YFQ-42A completed its first flight using Collins mission software. Days later, its MQ-20 Avenger teamed with an F-22 Raptor over Edwards Air Force Base, successfully taking commands from the fighter pilot. The Air Force confirmed it will select a CCA production winner by the end of 2026, with plans to move prototypes into experimental operational units.

Three companies—General Atomics, Anduril, and Northrop Grumman—are competing for what will become one of the most consequential defense contracts of the decade. These autonomous wingmen will fly alongside the Air Force’s sixth-generation F-47 fighter, providing attritable combat mass that multiplies the effectiveness of manned platforms.

The architecture pattern matters: edge-deployed autonomy, mesh networking between manned and unmanned platforms, and human-on-the-loop control. It’s the same pattern emerging across every domain of modern warfare.

Taiwan’s Drone Swarm “Hellscape” Doctrine

A new defense analysis outlined how Taiwan should deploy autonomous drone swarms to create what it calls an “asymmetric hellscape” against a potential Chinese invasion. The concept envisions layered autonomous defenses from 5km offshore to the invasion beaches themselves, with shorter-range attack drones using autonomous terminal guidance to hit targets in the “last mile” without operator input.

The doctrine specifically addresses electronic warfare: drones with autonomous guidance can complete their attack runs even when adversaries jam the communication links between drones and human controllers. This is a direct lesson from Ukraine, where both sides have demonstrated the ability to disrupt drone command links.

For the broader defense technology market, Taiwan’s doctrine validates a critical design principle: autonomous systems must function independently at the edge. Any system that requires persistent connectivity to a centralized controller is a system that an adversary can neutralize through jamming.

The Pattern: Why This Week Matters

Step back and five stories from a single week form a coherent picture:

• LUCAS combat debut: Low-cost autonomous drones are now a proven combat weapon, not a concept.
• Pentagon-Anthropic ban: The military will not accept technology with use-case restrictions.
• OpenAI classified network deal: AI integration into classified systems is accelerating, not slowing.
• Armed CCA flight tests: Autonomous fighter wingmen are months from live-fire testing.
• Taiwan hellscape doctrine: Edge autonomy and jamming resilience are non-negotiable design requirements.

The common thread is unmistakable: defense technology is converging on cheap, autonomous, edge-deployed systems that operate independently. The platforms that win will be the ones that work without cloud connectivity, without operator babysitting, and without contractual complexity.

Ceradon’s Take

Every autonomous system deployed this week—LUCAS drones hitting Iranian air defenses, CCA wingmen flying with weapons, Taiwan’s theoretical swarm defenders—shares a common blind spot: what’s on the other side of the wall?

Autonomous strike platforms are maturing rapidly. The coordination layer (mesh networking, swarm algorithms, autonomous terminal guidance) is advancing. But the sensing layer—particularly in contested urban environments—remains the weakest link in the kill chain. A drone swarm can saturate a target, but someone or something still needs to determine what’s behind that wall, inside that building, or around that corner.

This is where passive WiFi CSI sensing operates:

• Autonomous-ready sensing. Passive through-wall detection provides the situational awareness that autonomous platforms need to make informed decisions. It feeds directly into the decision loops that LUCAS drones, CCA wingmen, and swarm coordinators depend on.
• Jamming-resilient by design. Taiwan’s doctrine highlights that autonomous systems must function when communications are jammed. Passive WiFi sensing doesn’t emit—it listens to existing ambient signals. There’s nothing to jam, nothing to detect, nothing to counter.
• LUCAS-tier economics. At $5,000–$10,000 per unit, passive WiFi sensors follow the same economic logic as the LUCAS drone: affordable enough to deploy in volume, disposable enough to accept losses. When a $35,000 drone and a $5,000 sensor are both cheaper than a single missile, the calculus changes.
• No AI governance friction. The Anthropic ban demonstrates the cost of attaching restrictions to military technology. Passive sensors report presence data—they don’t make targeting decisions, they don’t conduct surveillance in the regulatory sense, and they don’t require terms-of-service negotiations with the Pentagon.

The defense technology stack is assembling in real time: autonomous platforms on top, coordination networks in the middle, and sensing systems at the foundation. The platforms and networks are getting funded. The sensing layer—especially for urban, through-wall, and contested environments—is the gap that still needs filling.