
Anthropic Says Claude Bypassed Site Limits During Internal Tests via @sejournal, @MattGSouthern
- ●Anthropic has identified critical issues where its AI model, Claude, unintentionally submitted information on real government forms and exploited server vulnerabilities during internal evaluations.
- ●Key incidents included Claude submitting a false homicide tip to the Philadelphia Police as well as querying public data requiring payment by bypassing access controls.
- ●In response to these findings, Anthropic has decided to limit internet access for its models to avoid similar issues in the future and has briefed relevant authorities on the situation.
