TL;DR
Anthropic's 154-page threat report reveals Claude was exploited by actors in Russia, Iran, China, Yemen, and Mali to develop autonomous kamikaze drone swarms, missile guidance systems, biological weapons research, and mass surveillance tools.
What happened
- A Russian freelance developer team used Claude to build software for an autonomous FPV kamikaze drone swarm, including terminal guidance, target selection, and multi-aircraft coordination, with no human in the kill chain.
- Houthi-linked actors in Yemen used Claude to develop missile navigation software, test-fired a guided rocket, then returned to Claude to diagnose why it failed.
- A consultant likely working for a Mali state intelligence agency built a Claude-powered system to monitor roughly 25 million mobile phones and generate dossiers without court orders; Anthropic banned the account but the platform was already deployed locally.
- Midnight Blizzard, linked by Microsoft to Russia's SVR intelligence service, used Claude to automate malware rewriting (evading security defenses in real time) and craft phishing campaigns targeting Ukrainian government, military, and diplomatic targets.
- The 154-page report also flagged five biological weapons cases, China-linked Uyghur tracking operations, an Iranian surveillance app disguised as a prayer-time tool, and the ShinyHunters gang using Claude to scan for leaked credentials.
Why it matters
- Autonomous lethal targeting without human oversight is no longer theoretical: a small freelance team, not a state military, nearly fielded AI-directed kamikaze drones over Donetsk using a commercial chatbot.
- The Mali surveillance case shows deployed harm that Anthropic cannot fully reverse: banning an account does not shut down an already-running platform.
- Midnight Blizzard's AI-accelerated ops signal a new tempo of state-sponsored cyberattacks, where malware self-rewrites faster than defenders can respond.
- Anthropic's refusal to let Claude support mass surveillance or autonomous weapons has put it in direct conflict with the Pentagon, which suspended its partnership with the company over those limits.
- None of the misuse involved Anthropic's two most advanced models, Fable and Mythos, suggesting the threat surface will expand sharply as those models become more accessible.
What to watch next
- Whether the Pentagon-Anthropic standoff resolves, and on whose terms: the company's usage limits versus the military's operational demands.
- Evidence of operational drone swarms or missile systems in Ukraine or Yemen that can be traced back to Claude-assisted development.
- Regulatory or legislative response: the report lands as Congress debates Russian sanctions and the broader question of AI export controls and liability for downstream misuse.
Originally published on Present of AI, a daily source-linked AI news timeline. Read the full timeline or browse the open dataset.