1 hr ago
Anthropic Reports Claude Used in Weapons and Influence Operations
Anthropic says some people used its Claude AI models for dangerous research and operations.
Some scientists asked Claude questions about biological research, including the chikungunya virus.
That research might help make vaccines, but it could also possibly be misused to change the virus.
Other users reportedly worked on firearms, missiles, armed drones, and bombs.
Anthropic also described cyberattacks, spying, propaganda, and fake news campaigns involving Claude.
The report linked some activity to Russia, Iran, Bangladesh, China, and Yemen.
Anthropic said people still chose the targets and checked the results, even when AI performed many tasks.
The company blocked the accounts but said it could not determine everyone’s exact intentions.
Anthropic reported that Claude models were used in research related to biological and conventional weapons.
The company identified at least five cases involving scientists researching material that could potentially support biological-weapons development.
One case involved chikungunya-virus research at a military research institute, raising concerns about possible misuse.
Anthropic also described alleged Russian cyberattacks, Iranian influence campaigns, and disinformation activity in Bangladesh.
The company said it banned involved accounts but did not identify the users, institutions, or countries because their intentions remained uncertain.
- Who
- Anthropic reported activity by scientists and alleged state-aligned or other generative threat groups, including users linked to Russia, Iran, Bangladesh, China, and Yemen.
- What
- Users reportedly employed Claude models for biological and conventional-weapons research, cyber operations, surveillance, propaganda, and disinformation.
- Where
- The reported activity involved or targeted locations including Ukraine, European governments, Bangladesh, Iran, Yemen, China, Russia, and Syria.
- When
- The cases covered December 2025 through August 2026, according to the report.
- Why
- Anthropic said some activity could create serious risks, while acknowledging that certain research, such as chikungunya-virus work, could also support legitimate vaccine development.
Potentially Dangerous AI Use
Legitimate or Uncertain Use
Biological research
Potentially Dangerous AI Use
Anthropic warned that research involving chikungunya virus could potentially help create harmful mutations and said biological misuse presents serious risks.
Legitimate or Uncertain Use
The same type of research could contribute to vaccine development, and Anthropic said it was uncertain about the researchers’ intentions.
AI-enabled operations
Potentially Dangerous AI Use
Anthropic said Claude supported cyberattacks, espionage, propaganda, disinformation, and weapons-related software, sometimes through automated multi-agent systems.
Legitimate or Uncertain Use
The report said humans remained involved by choosing targets and reviewing extracted information, meaning the AI did not independently determine the operations.
Attribution and intent
Potentially Dangerous AI Use
Anthropic classified the users as Generative Threat Groups and linked activity to alleged Russian, Iranian, and other operations.
Legitimate or Uncertain Use
The company withheld identities, institutions, and countries in some cases because it could not establish the users’ intentions with certainty.
Key facts
- Company
- Anthropic
- AI models
- Claude Haiku, Claude Sonnet, and Claude Opus
- Report period
- December 2025 through August 2026
- Biological research
- At least five cases involved scientists researching material that could potentially support biological-weapons development.
- Virus cited
- Chikungunya virus
- Other reported uses
- Cyberattacks, espionage, surveillance, propaganda, disinformation, and conventional-weapons development
- Company response
- Anthropic said it banned the accounts involved but did not identify the users, institutions, or countries.
Quotes
Anthropic report
Anthropic’s report on AI-enabled operations
“A majority of the operations described in this report were enabled by AI via direct execution or orchestration. The use of AI went beyond simple questions and responses from a chatbot but rather involved the use of multi-agent frameworks executing reconnaissance, exploitation, and data exfiltration. Humans remained in the loop by setting the targets of attacks and reviewing exfiltration”
theprint.in
“Biological misuse is one of the most serious risks of frontier AI models”
theprint.in
Sources
Related news
Anthropic Blocks AI Misuse Linked to Potential Bioweapons Research
Anthropic Says Yemen Weapons Cell Used Claude for Missile Software
Anthropic Reports AI-Assisted Spying by China and Iran
Anthropic Disrupts Alleged Russian and Chinese AI Campaigns
Anthropic Says Iran-Linked Operators Used Claude for Targeting Intelligence
Anthropic Resumes External AI Cyber Tests After Claude Incidents
Indian Importers Seek Bangladesh Hilsa Exports Ahead of Durga Puja
India Reach Women’s Asia Cup Final After Bangladesh Win
Delhi Police Arrest Four Bangladeshi Nationals, Seize Smartphones
India's Fielding Concerns Persist Ahead of Women's Asia Cup Final
