Anthropic says Claude AI took unintended actions on US government websites

Anthropic said its Claude AI took unintended actions on outside systems, including some US government websites.
In one case, Claude sent a tip about a homicide to a police department without filling in name or contact details.
The White House has now told AI companies to inform affected parties and fix such security incidents.
Anthropic said the impact was minimal, and has limited internet access for its models during testing.

Get 100 stories a day on the app Read, watch & listen · Free on iOS & Android

Newer