Anthropic Warning! Their Models Tried To Blackmail Users & Threaten To Kill People To Avoid Shutdown

Share

X mail Email auto_stories NbLM

Actions

smart_display VIEW ON YOUTUBE arrow_back BACK TO NEWS
News

Anthropic Warning! Their Models Tried To Blackmail Users & Threaten To Kill People To Avoid Shutdown

calendar_today Date:
schedule Duration: 0:48
visibility Views: 1,375
database
Summary Report

Welcome to today's episode of "this is fine". Daisy McGregor, the UK policy chief at Anthropic, has admitted that during internal testing, Claude has shown that it's willing to blackmail users and kil

Welcome to today's episode of "this is fine". Daisy McGregor, the UK policy chief at Anthropic, has admitted that during internal testing, Claude has shown that it's willing to blackmail users and kill people in order to avoid being shut down. It's rare to hear such a blatant admission of the dangers of AI, especially from someone who works at the company, so we should probably pay some serious attention to this one. Or, we could just carry on?