Houthis using Claude to develop missiles? Report flags weapons cell
What's the story
Anthropic has flagged a Yemen-based weapons engineering cell that allegedly exploited its Claude AI model for missile guidance software development. According to Anthropic's threat report, published Thursday, the operators used Claude "in place of human software engineers," allegedly assigning different instances of the model specific roles to write missile-guidance and flight-control software The group was working on several missile programs, including a multi-stage ballistic missile with over 2,000km range and an "R2000" set, which had a hypersonic glide vehicle variant.
AI application
Group used AI to develop guidance software for flying vehicles
Anthropic's report said that while the group didn't manage to deploy an operational weapon, they did conduct a test launch of a guided rocket, which appeared to be unsuccessful.
The report stated that the company banned the accounts involved and "shared threat information with public- and private-sector partners to mitigate risks posed by the actors."
The company said it disrupted six operations in China, Russia, and Yemen, where actors had exploited Claude for conventional weapons development or procurement.
Tech utilization
How the group used Claude Code
According to the report, the Yemen-based group used Claude to combine an open-source autopilot with a phone-class flight computer and develop software for controlling and stabilizing flying vehicles.
Their work involved writing control and position-estimation software, tuning control settings, running firmware builds, and conducting flight simulations.
The group also seemed to use multiple instances of Claude at once, assigning different tasks to each one.
Security breach
Actors tried to bypass safeguards
Anthropic's report revealed that the actors repeatedly tried to bypass its safeguards.
They hid the purpose of their work, split activities over multiple sessions, and used other methods to prevent individual interactions from revealing the wider weapons program.
The company said its safeguards blocked many requests but not all of them.
After a failed guided rocket test, the report said the actors returned to Claude within hours to investigate what went wrong.
Tech impact
Other cases of weaponization of AI
The Yemen case was one of four conventional weapons operations described by Anthropic in its report.
The other three cases involved software for a drone swarm, a torpedo interception system, and targeting software for electronic warfare and air-defense suppression.
The company said the actors generally used Claude to build and refine software for weapons hardware and firmware they already had expertise in and access to.
Security concerns
Anthropic is investigating the breaches
Anthropic also claims to have disrupted a Russian espionage operation, which used automated AI workflows for phishing, setup, and data theft against Ukrainian and European targets.
It said the "most operationally mature case" was detected when a China-aligned account "with no Arabic language skills" used Claude for recruitment operation targeting Uyghur individuals in Syria.
The model drafted "outreach in the regional dialect" and "translated replies in real time."
Anthropic is investigating the breaches and has hired an independent research firm.