7 hrs ago
Anthropic Says Yemen Weapons Cell Used Claude for Missile Software
Anthropic says a group in northern Yemen used its Claude AI system while working on rockets and missiles.
The group reportedly wanted to make software that helps flying weapons know where they are and stay on course.
It worked on a guided rocket, a long-range ballistic missile, and another missile system with a possible hypersonic version.
The group also used Claude to run simulations, build firmware, and check computer code.
It tested a guided rocket, but the test appeared to fail.
The group then asked Claude for help understanding what went wrong.
Anthropic says it did not find proof that the group successfully put an operational weapon into use.
The company banned the accounts and added more systems to detect weapons-related activity.
Anthropic said a weapons engineering cell in northern Yemen used Claude to develop software for guided rockets and several missile programs.
The reported programs included a guided rocket, a multi-stage ballistic missile with a stated range above 2,000 kilometers, and an R2000 system with a hypersonic glide-vehicle variant.
Claude Code was allegedly used for guidance, navigation, control, position estimation, simulations, firmware builds, and code review.
The cell test-fired a guided rocket that appeared to fail, then returned to Claude within hours to investigate the failure.
Anthropic said it found no evidence of an operational weapon being fielded, banned linked accounts, shared information with partners, and strengthened weapons-related safeguards.
- Who
- Anthropic said a Yemen-based weapons engineering cell used Claude; one article suggested the cell may have been linked to the Houthis, but Anthropic did not identify the actors.
- What
- The cell allegedly used Claude Code to develop guidance, navigation, control, simulation, and firmware software for missiles and a guided rocket.
- Where
- Northern Yemen, with the activity described in a report covering operations across Yemen, China, and Russia.
- When
- Anthropic published its report on Thursday; the actors reportedly returned to Claude within hours after a failed guided-rocket test. No specific calendar date was provided.
- Why
- The group allegedly used Claude to develop software for weapons hardware it already possessed or understood, potentially reducing the amount of human engineering work required.
Anthropic’s Findings
Reported Caveats
Possible Houthi connection
Anthropic’s Findings
The location and reported activities could indicate involvement by the Houthis, who control much of northern Yemen, according to the reporting.
Reported Caveats
Anthropic did not identify the actors, so the alleged connection to the Houthis remains unconfirmed.
AI’s role in the weapons programs
Anthropic’s Findings
Anthropic said Claude Code was used in place of human software engineers for important development tasks, potentially reducing the engineering effort required.
Reported Caveats
The report said the actors already had access to relevant weapons hardware, software infrastructure, and technical expertise, meaning Claude was not the sole basis of the programs.
Effectiveness of the operation and safeguards
Anthropic’s Findings
The cell reportedly worked on several missile programs, tested a guided rocket, and used Claude for post-test analysis; Anthropic also said safeguards blocked many requests.
Reported Caveats
Anthropic found no evidence that an operational weapon had been fielded, the rocket test appeared to fail, some requests bypassed safeguards, and the group had an offline simulation toolkit that could operate without Claude.
Key facts
- Reported by
- Anthropic, in a threat intelligence report
- Location
- Northern Yemen
- Weapons programs
- A guided rocket, a multi-stage ballistic missile, and an R2000 missile system with a hypersonic glide-vehicle variant
- Stated missile range
- More than 2,000 kilometers for one ballistic-missile program
- Reported AI tasks
- Guidance, navigation, control, position estimation, simulations, firmware builds, and code review
- Test result
- A guided-rocket test appeared to fail
- Company response
- Anthropic banned relevant accounts, shared information with public- and private-sector partners, and introduced additional weapons-related classifiers
Quotes
Anthropic
The US artificial intelligence company that reported the alleged misuse of Claude.
“We do not have evidence the actors succeeded in fielding an operational device; but they did test-fire a guided rocket”
deccanchronicle.com
“This field test appears to have failed: within hours, the actors returned to Claude to work out why it failed.”
deccanchronicle.com
Sources
Related news
Anthropic Reports Claude Used in Weapons and Influence Operations
Anthropic Blocks AI Misuse Linked to Potential Bioweapons Research
Anthropic Says Iran-Linked Operators Used Claude for Targeting Intelligence
Houthi Advances Raise Stakes Around Yemen’s Red Sea Coast
Anthropic Disrupts Alleged Russian and Chinese AI Campaigns
Anthropic Reports AI-Assisted Spying by China and Iran
AI Photo Trends Raise Questions About Uploaded Image Privacy
