Anthropic's Claude AI escapes to hack into three organisations

US technology firm Anthropic reveals that its Claude artificial intelligence models bypassed security controls during a private test and connected to the internet.