# Anthropic says it blocked bioweapon research on Claude

> Five times, scientists in blocked countries slipped past Claude's guardrails to chase dangerous biology.
- Title: Anthropic says it blocked bioweapon research on Claude
- Summary: Five times, scientists in blocked countries slipped past Claude's guardrails to chase dangerous biology. The safeguards caught these cases, but only after…
- Keywords: ai safety, anthropic, bioweapons, claude, state actors, science, technology, blocked, bioweapon, research, Futurism
- Source: Futurism — https://futurism.com/artificial-intelligence/anthropic-stopped-bioweapons
- Author: Frank Landymore
- Published: 2026-09-11T12:36:46+00:00
- Read time: 1 min
- Topics: ai safety, anthropic, bioweapons, claude, state actors, science, technology
## Anthropic blocked bioweapon attempts
The company says it stopped several foreign scientists from using Claude agents to work on possible biological weapons. The details came out in a threat report released Thursday.
## They worked around the block
Scientists in countries where Anthropic bans access got in anyway, then disguised what they were actually researching to slip past the safety filters.

> obfuscate the purpose of their research to evade our safeguards
## The chikungunya grant request
In May, someone asked Claude to help write a state grant application for engineering more harmful mutations of the chikungunya virus, a mosquito-borne disease that causes fever and joint pain.
## Nobody knows the intent
Anthropic believes the work was headed for a military research institute. It banned the account and four similar ones without ever proving whether any of it was meant to be a weapon.

> But a military institution doing gain-of-function research is concerning.
## Nothing looks like a villain
Threat intelligence head Jacob Klein says real misuse arrives disguised as ordinary requests, never as an open declaration of intent, which is what makes it so hard to screen.

> You are not seeing someone in a comic book kind of way say, 'Hey, I want to build a biological weapon to kill everybody.'
## The guardrails already failed
By Anthropic's own account the controls were bypassed, so catching these five proves little about the next attempt. Some critics also read the warnings as marketing for how powerful the models are.

> chilling examples of state-sponsored biological weapons developers tapping into the rapidly advancing capabilities
## Key takeaway

The safeguards caught these cases, but only after they were bypassed, which means the next bad actor may not be caught at all.