Anthropic has revealed some seriously disturbing examples of how people have attempted to misuse its Claude AI models.
The company says it has disrupted activity involving cyber-espionage, malware, propaganda and research that could potentially support biological weapons development.
One case involved requests connected to gain-of-function research involving the chikungunya virus.
Another investigation described actors believed to be linked to Yemen’s Houthis attempting to use Claude to help develop software for ballistic-missile guidance and control systems.
Anthropic says many harmful requests were blocked and associated accounts were banned, with information also shared with relevant partners and authorities.
We spent years asking whether AI could write our emails.
Apparently someone else was asking whether it could help build missile software.