Anthropic Reveals AI Models Targeted for Biological Weapons and Cyber Attacks

A new report published by Anthropic on Thursday details several alarming incidents where individuals attempted to use the company's artificial intelligence models to support biological weapons development, alongside other malicious activities. The release of this report follows growing public safety concerns and debates over whether increasingly powerful AI models pose significant risks.

The report outlines how users tried to misuse the Claude tool for cyber attacks, influence operations, surveillance, biology, and weapons manufacturing, and explains how Anthropic successfully identified and stopped them. Among the highlights was a specific prompt where a user sought assistance with a grant application to research the chikungunya virus, a mosquito-borne illness with no licensed therapeutic that causes severe, prolonged symptoms.

Anthropic noted that internal checks and workflow improvements allowed them to ban associated accounts and integrate investigative findings into new preventative measures. While striving to maintain usefulness for legitimate scientific research, the company faces mounting public scrutiny, legislative proposals in the US Congress, and political pressure from figures like US senator Bernie Sanders regarding Big Tech accountability.