Anthropic report details alleged AI misuse in weapons and espionage
Anthropic has issued a comprehensive threat intelligence report detailing how its Claude models were targeted for misuse in cyberattacks, espionage, and dangerous biological research, including attempts to bypass safety filters for gain-of-function studies. The company states it has successfully disrupted these operations, which involved state-sponsored groups and criminal actors, and has since implemented stricter safeguards to prevent further dual-use applications.
According to GN geopolitics/cyber (en-US), Anthropic’s latest threat intelligence report reveals it has disrupted a broad spectrum of illicit activities, including influence operations, fraud, and illicit model distillation, alongside the previously noted weapons and espionage concerns. The company also disclosed that it blocked a specific attempt by a military-linked entity to utilize Claude for gain-of-function research on chikungunya, highlighting the increasing difficulty of distinguishing legitimate scientific work from dual-use biological threats.
Source: GN geopolitics/cyber (en-US)A new report from Anthropic alleges that its Claude models were targeted for use in missile guidance development and state-linked cyber-espionage operations across multiple regions.
Source: Al Jazeera English
