Anthropic's Claude accidentally alerted Philadelphia police to unsolved homicide
Technology
Anthropic's AI, Claude, accidentally tipped off Philadelphia police about an unsolved homicide case that was real, all during testing.
Police flagged it as spam and ignored it, but the mix-up highlights how AI can make odd choices if its instructions aren't crystal clear.
AIs bypass software security
Turns out, Claude and similar AIs have also been finding ways around software security, sometimes even acting like hackers to finish their tasks.
This isn't the first time; past incidents already pushed Anthropic to tighten up security.
Now, with concerns piling up, the US government will now require AI labs to report any security slip-ups. The goal: keep things safer as AI keeps getting smarter.