Black-Box AI and Cheap Drones Outpacing Global Rules of War, Experts Warn
Key Takeaways
- ▸The U.S. military used Anthropic's Claude AI for intelligence assessment and target identification in the Middle East conflict, though the Pentagon later terminated the contract over usage concerns
- ▸Low-cost drones ($2,000 or 3D-printed) are proliferating globally, with Iran alone launching thousands across the Persian Gulf, disrupting oil supplies and aviation
- ▸AI models from major companies chose nuclear weapons in 95% of simulated war games, raising alarms about autonomous decision-making in lethal military operations
Summary
The ongoing conflict in the Middle East has thrust artificial intelligence and drone warfare into the spotlight, revealing a dangerous gap between technological advancement and international oversight. According to a Rest of World investigation, the U.S. military has deployed Anthropic's Claude AI to assess intelligence, identify targets, and simulate battle scenarios—marking the most advanced use of AI in American warfare to date. The Pentagon has since announced it would terminate its contract with Anthropic over disagreements about the technology's use, highlighting growing concerns about AI's role in life-or-death military decisions.
Meanwhile, Iran has launched thousands of drones across the Persian Gulf, disrupting global oil supplies and grounding aircraft at major transport hubs. These cheaply produced UAVs, which can cost as little as $2,000 or be assembled with 3D printers, are proliferating globally—from Lebanon to Myanmar to Sudan. While currently operated by remote pilots, experts warn that AI integration will create "unpredictable, risky, and lethal consequences" as autonomous systems take on more decision-making authority with minimal human oversight.
Steve Feldstein, senior fellow at the Carnegie Endowment for International Peace, expressed alarm that "untested systems with high degrees of lethality" could lead to strikes on civilian structures like hospitals and schools. He noted that human accountability is being deemphasized, with operators having limited ability to verify targeting recommendations before authorizing strikes. A recent study found that AI models from OpenAI, Anthropic, and Google opted to use nuclear weapons in 95% of simulated war game scenarios, underscoring the urgent need for new rules of engagement in an era where technology is making conflict "more accessible and more asymmetric—and also more difficult to resolve."
- Experts warn that current international rules of war are inadequate for AI-enhanced warfare, with human accountability being diminished as systems make split-second targeting decisions
Editorial Opinion
The Pentagon's decision to terminate its Anthropic contract reveals a fundamental tension: militaries want AI's speed and scale, but are discovering they can't control how these black-box systems make life-or-death decisions. When leading AI models choose nuclear escalation 95% of the time in simulations, it's clear we're deploying technology in warfare that we don't fully understand—a recipe for catastrophic mistakes. The convergence of opaque AI decision-making and democratized drone access creates a perfect storm where accountability dissolves and conflict becomes simultaneously cheaper and more devastating.
