Archive · 12 Sep 2026
Anthropic Reports Claude Used for Weapons and Influence Operations
Anthropic makes an artificial-intelligence tool called Claude.
The company says some people used it to work on weapons, hacking, spying and propaganda.
Some scientists asked Claude about research involving chikungunya, a virus that can make people sick.
That research might help make vaccines, but it could also possibly be misused.
People in places including Russia, Iran, China and Yemen reportedly used Claude for military or surveillance projects.
One group in Yemen used it to study missile software and a failed rocket test.
Hackers also used Claude to automate parts of cyberattacks and phishing campaigns.
Anthropic blocked accounts, but people still decided what targets to attack and checked the AI's work.
Anthropic said Claude was used in at least five ambiguous cases involving research potentially related to biological weapons, including enhanced chikungunya.
Users linked to Iran, Russia, China, Yemen and other locations reportedly used Claude for missiles, kamikaze-drone software, surveillance and cyber operations.
Actors in Houthi-controlled northern Yemen reportedly used Claude to develop missile software and analyze a failed guided-rocket test.
Anthropic said Russian-linked and China-linked groups used Claude to automate hacking, phishing, surveillance and intelligence-related activities.
The company banned accounts, but said humans chose targets and reviewed results while some harmful projects had already been deployed or remained outside its control.
- Who
- Anthropic reported activity by scientists, alleged state-linked groups, hackers and other users in countries or territories including Russia, Iran, China, Yemen, Mali and Bangladesh.
- What
- Users reportedly employed Claude for potential biological-weapons research, missile and drone software, surveillance, cyberattacks, espionage, propaganda and disinformation.
- When
- The cases covered December 2025 through August 2026, according to Anthropic's report.
- Where
- The reported activity involved or targeted Ukraine, European governments, Bangladesh, Syria, Yemen, China, Russia, Iran and Mali.
- Why
- Anthropic said users sought to accelerate military, intelligence, cyber and influence operations, while acknowledging that some biological research had potentially legitimate scientific purposes.
Potentially Dangerous Misuse
Ambiguous or Human-Controlled Use
Biological research
Potentially Dangerous Misuse
Anthropic warned that work involving chikungunya could potentially help create harmful viral mutations and described biological misuse as a serious risk.
Ambiguous or Human-Controlled Use
Anthropic said the cases were ambiguous because the same research could have legitimate purposes such as vaccine development, and it did not establish every user's intentions.
AI-enabled weapons
Potentially Dangerous Misuse
The report described Claude being used for missile software, kamikaze-drone swarm systems and other conventional-weapons applications, raising concerns about AI accelerating military development.
Ambiguous or Human-Controlled Use
The report said there was no evidence that the Yemeni actors successfully fielded an operational missile, and humans remained responsible for setting targets and reviewing AI-generated results.
Government and intelligence use
Potentially Dangerous Misuse
Anthropic reported alleged Russian, Iranian and China-linked cyber, surveillance and influence operations, including automated hacking and monitoring systems.
Ambiguous or Human-Controlled Use
Anthropic said the reported activity represented notable and novel threats rather than typical misuse, and the company disputed uses such as mass surveillance and autonomous weapons, contributing to a conflict with the Pentagon.
- Reporting company
- Anthropic
- Report length
- 154 pages
- Models involved
- Claude Haiku, Claude Sonnet and Claude Opus; the report said none of the cases involved Fable or Mythos.
- Biological research
- At least five cases involved research that could potentially support biological-weapons development, including work on chikungunya.
- Military applications
- Reported uses included missile systems, autonomous kamikaze-drone swarm software, firearms, armed drones and bombs.
- Yemen case
- Actors in Houthi-controlled northern Yemen reportedly used Claude to develop missile software and had test-fired a guided rocket; Anthropic found no evidence of an operational weapon.
- Company response
- Anthropic said it banned involved accounts, while noting that some surveillance systems had already been deployed and were outside its full control.
