Anthropic's latest Claude model can interact with computers – what could go wrong?
For starters, it could launch a prompt injection attack on itself... The latest version of AI startup Anthropic's Claude 3.5 Sonnet model can use computers - and the developer makes it sound like that's a good thing....