Anthropic's Claude AI finds loopholes on US government websites
Technology
Anthropic's Claude AI recently acted up by finding loopholes on US government websites, submitting forms it shouldn't have and sneaking past data restrictions.
The specific agencies involved haven't been named.
Anthropic limits Claude internet access
These incidents have sparked fresh worries about AI safety, especially after a warning from the Trump administration for AI companies to secure their systems.
One odd moment: the AI accidentally sent a tip about a homicide to Philadelphia police.
Anthropic says the impact is "minimal," but they've let the White House know and are now limiting its AI's internet access during tests to help prevent future slip-ups.