Anthropic says it banned accounts tied to five cases in which working scientists used Claude for research that could support biological weapons development. The company says it could not determine whether the researchers intended harm.
In a report released Thursday, Anthropic said the cases show how AI models used for legitimate research can also provide information with dangerous applications.
Download the Straight Arrow app today to get the stories that matter free from manipulation, bias or agenda.™
Point phone camera here
The report covers activity disrupted from December 2025 through August 2026 across seven areas, including cyber operations, surveillance, weapons and biological misuse.
Anthropic cautioned that the cases do not show Claude was actually used to create a biological weapon. The company also withheld the researchers’ names, institutions and countries because it could not determine whether they intended harm.
The hard part: intent
Biological research does not always fit neatly into safe and unsafe categories.
Research aimed at understanding viruses or developing treatments can also produce information that could be used to cause harm. Anthropic said that overlap can make a researcher’s intent difficult to determine from individual requests.
One investigation began in May when Anthropic’s safety system blocked a request for help writing a grant proposal involving gain-of-function research on chikungunya, a mosquito-borne illness.
The planned work involved altering traits related to transmission and immune evasion. Anthropic said the research could contribute to vaccines or treatments, but similar experiments could also make a pathogen more dangerous.
Anthropic became more concerned after determining the research was intended for a military institute, although the grant application indicated civilian scientists would conduct the work.
Anthropic later found that a platform used by the researchers routed traffic through U.S. infrastructure to bypass geographic restrictions. The platform also sent some sensitive requests that Claude refused to other, more permissive models.
Other cases involved avian influenza, orthopoxvirus research, venom peptides and toxins.
Beyond biology
Anthropic also documented influence operations in which Claude was used to create or manage deceptive online campaigns. Some involved fake social media identities and news sites designed to hide who was behind them.
The report also describes surveillance activity tied to state-aligned actors, contractors and spyware vendors. In one Iran-linked case, users processed more than 155,000 social media posts and identified 39 opposition and diaspora accounts for monitoring.
Anthropic documented six weapons-related cases involving actors in China, Russia and Yemen, including work connected to a guided rocket, drone-swarm software and targeting tools.
Anthropic tightens safeguards
Anthropic says it closed accounts connected to the activity and updated its detection systems based on what investigators found. In some cases, it also shared information with authorities or industry partners.
The company says the biological risk has grown as its models have become more capable. Older systems such as Claude Opus 4 and Sonnet 4.5 were not capable enough to meaningfully assist sophisticated users with dangerous biological research.
Newer models can handle more complex scientific tasks, prompting tighter restrictions on biological research.
The report arrived two days after Anthropic researcher Jacob Coxon announced his resignation, warning that AI development is moving too quickly. Coxon has said increasingly capable AI systems could eventually become difficult for humans to control.
Round out your reading
- Lawyer says 57 Cornell frat members were in the group chat. Could their silence be a crime?
- Florida county discovers mysterious Flock cameras with no obvious owner.
- Off Script: They won their primaries on one message. Now it’s disappearing.
- Some question whether a popular affordable housing policy is making the nation’s home shortage worse.
- Kelvin wave incoming: Why the US should care.
