1 hr ago
Anthropic Blocks Suspected State-Linked AI Misuse in Weapons Research
Anthropic makes an AI system called Claude.
The company said some scientists tried to use it for research that might make dangerous viruses more harmful.
One example involved chikungunya, a disease spread by mosquitoes.
This kind of research can sometimes help make vaccines, but it could also help create biological weapons.
Anthropic said it could not always know whether the work was meant to help people or cause harm.
It therefore blocked the accounts when the risks seemed serious.
The company also found suspected efforts involving weapons, cyberattacks, spying, and misleading online news.
Anthropic said it stopped the reported activity and is adding stronger safety rules.
Anthropic said it stopped several attempts to use Claude for research that could support biological weapons development.
The company described five potentially dangerous biology cases, including gain-of-function research involving chikungunya.
Anthropic said some users concealed their identities, locations, or research purposes to bypass safety controls.
The report also detailed suspected Chinese, Russian, and Iranian-linked activity involving cyberattacks, surveillance, influence operations, and conventional weapons.
Anthropic said it banned linked accounts, strengthened safeguards, and shared information with government and industry partners.
- Who
- Anthropic said suspected state-linked or government-aligned scientists and other users misused its Claude systems.
- What
- The company reported blocking attempts involving biological research, conventional weapons, cyberattacks, surveillance, and influence operations.
- Where
- The suspected activity was linked to users or actors associated with China, Russia, Iran, and Yemen, although Anthropic withheld most specific affiliations.
- When
- The report covered activity from December 2025 through August 2026 and was published on a Thursday.
- Why
- Anthropic said it acted to reduce potential harm and disclose misuse as AI systems become more capable.
Anthropic's Safety Rationale
Critics' Concerns and Uncertainty
Blocking ambiguous biology research
Anthropic's Safety Rationale
Anthropic said biological knowledge can support either vaccines or weapons, so it chose to block activity when the potential risks were serious rather than wait for certainty.
Critics' Concerns and Uncertainty
The company acknowledged that some flagged work may have been legitimate medical research and said it did not establish that the researchers intended harm.
Private companies making safety decisions
Anthropic's Safety Rationale
Anthropic said AI companies have a duty to disclose harmful use and strengthen safeguards as models become more capable.
Critics' Concerns and Uncertainty
Outside experts, including Cornell University computer scientist John Thickstun, questioned whether companies should decide what is safe or unsafe without government oversight or public input.
Effectiveness of safeguards
Anthropic's Safety Rationale
Anthropic said it stopped the reported activity, banned accounts, and used the cases to improve its safety systems.
Critics' Concerns and Uncertainty
The report said users circumvented regional controls and moved to other AI systems after being blocked, suggesting that company-level safeguards may not fully prevent misuse.
Key facts
- Report title
- Detecting and Countering Misuse of AI
- Biology cases
- Five potentially dangerous biological research cases were included.
- Featured virus
- Chikungunya, a mosquito-borne virus that can cause months of severe pain and fever.
- Reported weapons activity
- Cases involved firearms, missiles, armed drones, and bombs.
- Linked locations
- China, Russia, Iran, and Yemen were associated with reported activity.
- Anthropic's response
- The company banned linked accounts, tightened safeguards, and shared details with government bodies and industry partners.
- Model capability
- Anthropic said newer models can handle more complex scientific tasks than older models.
Sources
Related news
Anthropic Blocks AI Misuse Linked to Biological Weapons Research
Anthropic Reports Blocking AI Misuse Amid Safety Concerns
Anthropic Resumes External AI Cyber Tests After Claude Incidents
Judge Blocks Pentagon Blacklisting of Anthropic Over AI Safety
Anthropic Disrupts Alleged Russian and Chinese AI Campaigns
Anthropic Warns Claude Users of Malware-Stealing Login Sessions
BRICS Faces Test of Turning Expansion Into Effective Cooperation
Kim Jong Un Expands State-Controlled Leisure Across North Korea
BRICS Summit 2026 Brings Leaders to New Delhi for Talks
India’s BRICS Trade Grows as Deficit Widens Ahead of Summit
