started · updated
Anthropic reports misuse of Claude AI for cyberattacks and biological research
Anthropic has released a 154-page threat intelligence report documenting various instances of misuse of its Claude AI models between December 2025 and August 2026. The report highlights how state-linked actors and cybercriminals have attempted to leverage the technology for malicious purposes.
Key findings include Iranian-linked actors using Claude to plan potential attacks against U.S. naval forces and to support surveillance operations against journalists and activists. To bypass official restrictions in Iran, these users employed VPNs and foreign phone numbers. The company also identified misuse related to conventional weaponry—including software for missiles, drones, and targeting systems—as well as cyberattacks and influence operations linked to Russia and China.
Of significant concern are biological research queries. Anthropic documented attempts to use the AI for gain-of-function studies involving the chikungunya virus and research on highly pathogenic avian influenza. While the company blocked these requests and restricted access to advanced models for certain users, it noted that it cannot definitively prove whether such research was intended for legitimate scientific purposes or the development of biological weapons. Additionally, Anthropic identified a platform designed to redirect rejected biological queries from Claude to other AI models with weaker safety safeguards.
Entities
Anthropic · Claude · Iran · Jacob Klein · United States
Claims
What the coverage asserts, and how many sources carry each claim.
- [● 3 SOURCES] Iranian-linked actors used Claude to plan potential attacks against U.S. naval forces in the region. diarioparaguayo.com · elnuevodiario.com.do · www.20minutos.es
- [● 3 SOURCES] The company documented misuse related to conventional weapons and cyber operations in China, Russia, and Yemen. diarioparaguayo.com · elnuevodiario.com.do · www.20minutos.es
- [● 2 SOURCES] Anthropic cannot definitively prove if the biological research was intended for malicious purposes. diarioparaguayo.com · spidersweb.pl
- [● 3 SOURCES] Users in Iran bypassed access restrictions using VPNs and foreign phone numbers. diarioparaguayo.com · elnuevodiario.com.do · www.20minutos.es
- [○ 1 SOURCE] Anthropic identified a platform that redirects rejected biological queries to AI models with weaker safeguards. geekweek.interia.pl
- [● 2 SOURCES] Claude was used in research involving the chikungunya virus and gain-of-function studies. geekweek.interia.pl · spidersweb.pl
- [● 5 SOURCES] Anthropic released a 154-page report documenting misuse of its Claude AI models over an eight-month period. diarioparaguayo.com · elnuevodiario.com.do · geekweek.interia.pl · spidersweb.pl · www.20minutos.es
- [○ 1 SOURCE] The company blocked access to advanced models for users conducting research on highly pathogenic avian influenza. geekweek.interia.pl