Back
Automatically translated version. May contain inaccuracies compared to the original.
The company also found Claude being used for propaganda, mass surveillance, and targeting.
Anthropic said on Thursday that it had blocked scientists who used its AI models, Claude, “in ways that could contribute to the development of biological weapons,” CBS News reports.
The company disclosed this in a major report, which also described other cases of harmful AI use related to surveillance, fraud, weapon development, and propaganda.
According to Anthropic, the cases cited in the report represent examples of the “most notable and new threatening activity” the company has detected to date. The threats involved include criminals, likely state-backed groups, spyware vendors, and state propaganda entities.
Anthropic described five instances of activity that could potentially facilitate the development of biological weapons. The company called the misuse of AI in the biological domain “one of the most serious risks of advanced AI models.”
At the same time, the company stressed that the identified cases “demonstrate that the relevant capabilities exist” in AI models, but “do not prove that such capabilities will ever be used to create biological weapons in the real world.”
The individuals involved in these cases are professional scientists. Anthropic chose not to disclose their names.
The company also emphasized that it cannot claim those people intended to cause harm.
“When we discovered and investigated these cases, we blocked user accounts and incorporated the investigation results into our safety systems, guardrails, and threat- analysis processes to better prevent, detect, and halt such activity in the future,” the report states.
Newer and more advanced models, such as Claude Fable 5, have stronger safeguards compared with earlier models and restrict access to “a broad range of queries related to dual-use biological research,” Anthropic notes.
The company emphasizes that AI capabilities in biology can be used for both beneficial and harmful purposes.
“The same information that could be used to develop biological weapons could also be used, for example, to create vaccines or medicines against disease,” Anthropic explained.
Propaganda and surveillance
Anthropic also said it had identified and removed accounts that used Claude to conduct influence operations aimed at shaping public opinion. Among them were three accounts linked to Iranian state actors.
According to the company, in each case the operation was conducted by a person working for or on behalf of an Iranian state propaganda organization.
Claude was used to create content, cloak posts as independent news materials, and spread content on social networks, including X, Instagram, and TikTok.
One Iran-linked threat actor used Claude to develop recommendations on selecting targets among the US Navy in the region, Anthropic reported.
The same account used AI to create software for an internal mass-surveillance platform intended for Iranian government systems.
Claude was also used to create and run surveillance operations by subjects in China and West Africa, and in some cases to identify targets.
Anthropic said it found cases where AI was actually used in place of an engineer's command.
In China, three accounts tied to PRC municipal security services used Claude for surveillance and transnational repression. Among them was a municipal department that compiled profiles of activists and organizations abroad.
Document: PDF proof of the original version of the news item "Claude використовували у розробці потенційної біозброї — Anthropic заблокувала користувачів". It records the publication content at the moment of the first scan, the preservation date and the source: ZN.ua.
