Archive · 11 Sep 2026
Anthropic Blocks AI Misuse Linked to Potential Bioweapons Research
Anthropic makes AI models called Claude.
The company says some people asked Claude for help with biological research that could make a virus more dangerous.
One example involved chikungunya, a virus spread by mosquitoes.
The proposed research included changing the virus and testing versions in animals to find more harmful ones.
This kind of research might help make vaccines, but it could also make a pathogen more dangerous.
Anthropic said the work appeared connected to a military research institute, but it did not prove that the researchers wanted to make weapons.
The company blocked the accounts and added stronger safety rules to newer models.
Anthropic also found Claude being used for cyberattacks, spying, propaganda, scams, and conventional-weapons work.
Some experts said governments should help set these rules instead of leaving major decisions entirely to AI companies.
Anthropic reported five biological-misuse cases detected and disrupted between December 2025 and August 2026.
One case involved a grant proposal for chikungunya research seeking mutations affecting transmissibility, immune evasion, and harmful effects.
Anthropic said the work appeared connected to a military research institute but could not establish that the scientists intended to develop weapons.
The report also described misuse involving cyberattacks, surveillance, propaganda, scams, conventional weapons, and unauthorized model copying.
Anthropic banned linked accounts, strengthened safeguards for newer models, and shared information with authorities and industry partners.
- Who
- Anthropic, Claude users including scientists and suspected state-linked or criminal actors, and outside experts including John Thickstun.
- What
- Anthropic reported blocking attempts to misuse its AI for potentially dangerous biological research and other harmful activities.
- When
- The activity was detected between December 2025 and August 2026; Anthropic published its report on September 11, 2026.
- Where
- The reported activity involved users or operations linked to China, Iran, Russia, Turkey, Yemen, the Persian Gulf, South Asia, Africa, and Europe; most biological-case locations were withheld.
- Why
- Anthropic said increasingly capable models could assist with complex scientific and harmful tasks, so it strengthened safeguards and acted cautiously where legitimate and dangerous research overlapped.
Stronger AI safeguards
Broader research access and public oversight
Dual-use biological research
Stronger AI safeguards
Anthropic said caution was necessary because advanced AI could assist with research that makes pathogens more dangerous, and the consequences of missing malicious activity could be severe.
Broader research access and public oversight
The same biological knowledge can support legitimate work such as vaccines and treatments, and Anthropic said it could not establish that the scientists intended harm.
Company-led enforcement
Stronger AI safeguards
Anthropic blocked accounts, restricted suspicious requests, and expanded controls on dual-use biological research as newer models became more capable.
Broader research access and public oversight
John Thickstun said it is uncomfortable for companies to make society-wide judgments about safe and unsafe behavior without democratic or deliberative government oversight.
Public disclosure
Stronger AI safeguards
Anthropic said publishing details could help governments, civil society, and other AI developers recognize emerging misuse and improve collective defenses.
Broader research access and public oversight
The company withheld names, institutions, countries, pathogens, and techniques in most biological cases because identifying researchers or laboratories could expose them to harm.
- Company
- Anthropic
- AI system
- Claude
- Report period
- December 2025 through August 2026
- Biological cases
- Five cases involving research that could potentially support biological-weapons development
- Highlighted case
- A chikungunya gain-of-function proposal involving mutations related to transmissibility, immune evasion, and harmful effects
- Other misuse
- Cyber operations, influence operations, surveillance, scams and fraud, conventional weapons development, and model distillation
- Response
- Anthropic banned linked accounts, strengthened safeguards, and shared intelligence with authorities and industry partners
Sources
Scientists Tried To Use AI Models For Potential Biological Weapons Research: Anthropic
Anthropic Says It Blocked Misuse of Its AI That Could Have Supported Biological Weapons
Anthropic says It disrupted attempts to misuse AI for biological weapons research
Anthropic Says It Blocked AI Misuse After Its Whistleblower's Safety Warning
Anthropic blocks weapons work queries on Claude AI from China, Russia users
'Malicious Misuse': Anthropic Says It Blocked Scientists Who Used AI For Bioweapons Research
Anthropic says it blocked misuse of its AI that could have supported biological weapons
Anthropic Says It Blocked State-Linked Scientists From Using Claude For Bioweapons Research
