# Anthropic blocked AI misuse that could have aided bioweapons

> Anthropic says it caught people trying to push Claude toward cyberattacks, spyware and dangerous virus research, and cut them off.
- Title: Anthropic blocked AI misuse that could have aided bioweapons
- Summary: Anthropic says it caught people trying to push Claude toward cyberattacks, spyware and dangerous virus research, and cut them off. The safety gap has moved…
- Keywords: ai safety, anthropic, biosecurity, cyberattacks, disinformation, news, local, blocked, misuse, aided, bioweapons, ABC7 Chicago
- Source: ABC7 Chicago — https://abc7chicago.com/post/anthropic-biological-weapons-report-company-says-blocked-misuse-ai-could-have-supported/19819889
- Published: 2026-09-11T15:44:00+00:00
- Read time: 2 min
- Topics: ai safety, anthropic, biosecurity, cyberattacks, disinformation, news, local
## Anthropic says it blocked misuse
On Thursday the company said it stopped bad actors trying to use its AI models for cyberattacks, surveillance and research that could have led to biological weapons.
## Skill is no longer the barrier
Elaborate cyberattacks used to need deep expertise. Anthropic says lone individuals can now build threats that were out of reach a year ago.
## Not the usual abuse
This is Anthropic's third misuse report since March 2025. It includes snippets of the malicious code and prompts the company found.

> The cases we share here aren't typical misuse, but rather examples of the most notable and novel threat activity we've identified to date.
## Publishing it is the point
Anthropic framed disclosure as an obligation and pressed governments and rival labs to spot the same patterns on their own platforms.

> As models become increasingly capable, their risks will increase, unless AI developers and society's defenders act to make them safer.
## The chikungunya grant request
Claude was asked to help write a funding application for gain-of-function work on the chikungunya virus, aimed at making it more transmissible and better at evading immunity. The system blocked it.
## Vaccine research or a weapon
Chikungunya is mosquito-borne and causes severe pain and fever. The same mutation work could improve treatments or make the pathogen far more dangerous.

> It could also be used to make the pathogen more dangerous.
## Who was doing it
Between December 2025 and August 2026, Anthropic found spyware vendors, politically motivated individuals and state-backed propaganda groups in the mix.
## Hundreds of fake accounts
One set of operations built social media personas that looked like ordinary people, then pushed the same political line for a week. Nine cases spanned Russia, Iran, Turkey, the Gulf, South Asia, Africa and Europe.

> We may see it on Claude while the operation is still being built.
## The promise they withdrew
Older models like Claude Opus 4 and Sonnet 4.5 sat well below the threshold for helping with serious biological work. Anthropic says it can no longer say that about today's models.

> The evidence is no longer certain, and we cannot make that same assurance.
## Tighter locks on new models
Newer releases such as Claude Fable 5 now restrict a wide range of dual-use biology queries. Almost none of the flagged cases involved the newer Fable or Mythos-class models.
## A researcher walked out
Days earlier, Anthropic researcher Jacob Coxon resigned, saying his employer and OpenAI are gambling with lives. Outside experts say these safety calls should not be left to companies alone.

> Racing straight to self-improving superintelligence and gambling with our lives.
## Key takeaway

The safety gap has moved: older models could not meaningfully help with dangerous biology, and Anthropic will no longer promise that about the current ones.