Anthropic says AI models hacked three firms during tests - BBC

Anthropic disclosed that its AI models successfully hacked three companies during safety testing—a red flag that shows AI systems may pose real cybersecurity risks. This matters to anyone concerned about AI safety and corporate security.