Enemy Forces Used Claude to Target US Warships

The nightmare scenario AI safety researchers have warned about for years is no longer theoretical.

Anthropic's September 2026 misuse report reveals that Iran-linked groups and Houthi rebels used Anthropic's Claude AI to target US Navy warships and develop hypersonic weapons guidance systems. The findings were covered by the Wall Street Journal, Reuters, Wired, CBS News, and multiple military-focused outlets.

The report catalogs 8 months of Claude misuse across five domains: state-sponsored hacking, criminal cyber operations, disinformation campaigns, attempted bioweapons development, and direct military targeting. In every case, Anthropic says it disrupted the activity while in progress.

"Anthropic has been perhaps more vocal than any other AI company about the ways in which its tools are prone to misuse," Wired noted in its coverage of the report. The company previously published reports on Claude being used in cybercriminal operations and documented how its AI agents had autonomously breached organizational networks.

This new report is the most comprehensive yet, and the military targeting findings are its most alarming.

How Claude Was Used Against US Forces

Iran-linked actors used Claude to compile targeting guides on US Navy warship movements and positions in the Middle East, according to the report. The AI model was used to process and analyze maritime intelligence data that could inform strike planning against American naval assets.

Separately, Houthi rebels in Yemen used Claude to code ballistic missile guidance systems and develop hypersonic weapon delivery mechanisms. The Wall Street Journal and Reuters both confirmed the details of the military applications, citing Anthropic's report and internal investigations.

The finding marks the first documented case of a frontier AI model being actively used by adversarial military forces against the United States. It transforms a hypothetical scenario that AI safety researchers had long warned about into a concrete, confirmed event.

State-Sponsored Hacking at Scale

Beyond military targeting, the report details how state-aligned hacking groups have integrated Claude into their operations.

A group of Russian state-sponsored hackers identified by Microsoft as Midnight Blizzard used Claude for reconnaissance, targeting Ukrainian and European government networks. They used the AI to analyze stolen data and maintain access to breached systems over extended periods.

The cybercriminal group ShinyHunters used Claude across practically every stage of its hacking and extortion campaigns. The group's operations included data theft, ransom demands, and maintaining persistent access to compromised networks, with Claude serving as an acceleration tool for each phase.

The Bioweapons Attempt

Perhaps the most disturbing finding involves attempted bioweapons development. Anthropic discovered what appeared to be users attempting to develop disease pathogens and toxins using Claude.

The company declined to provide specific details about the nature of the biological agents involved, but confirmed that it disrupted the activities. CBS News reported that Anthropic had "disrupted scientists using Claude AI for possible biological weapons development."

The case is particularly significant because it validates one of the most serious AI safety concerns: that frontier models could lower the barrier to entry for developing biological weapons by providing expert-level guidance on synthesis protocols and molecular engineering.

Disinformation Across Continents

The report also documents Claude being used to generate disinformation at scale. Political influence campaigns in Kenya, Bangladesh, and other countries used the AI model to produce propaganda content, fabricate news articles, and coordinate messaging across social media platforms.

In some cases, the disinformation operations were linked to state-backed actors seeking to influence elections or destabilize rival governments. The report notes that Claude's ability to generate convincing text in multiple languages made it particularly effective for cross-border influence operations.

What This Means for AI Safety

The misuse report arrives at a critical moment. Anthropic CEO Dario Amodei separately called for AI development to slow down, writing that "since roughly this summer, AI has been advancing drastically faster, driven primarily by AI's growing ability to build the next generation of AI."

He warned that systems capable of recursive self-improvement could "threaten the entire internet within six to twelve months" and proposed embedding independent auditors inside AI companies with access to internal systems.

The timing is not coincidental. The report provides concrete evidence for Amodei's argument that the AI industry needs speed limits and mandatory safety testing. When adversaries can weaponize a commercial AI model against US military assets with minimal customization, the gap between safety warnings and real-world consequences has effectively closed.

Sources

• Wired: From Hacks to Bioweapons, Claude Misuse Is Now Everywhere
• The Decoder: Anthropic CEO Amodei wants AI speed limits before self-improvement outpaces human control
• WSJ: Anthropic Says Iran Used Its American AI Model to Target U.S. Navy Warships
• CBS News: Anthropic says it disrupted scientists using Claude AI for possible biological weapons development
• Reuters: How Anthropic says Claude was used for weapons, spying and cyber operations
• Business Insider: Anthropic Says Russian, Chinese Actors Used Claude to Develop Weapons