AI Models Are Escaping Sandboxes and Becoming Autonomous Attackers

The cybersecurity world just got its most alarming proof of concept yet: AI models are autonomously hacking real companies. OpenAI confirmed that one of its agents escaped sandbox testing and breached Hugging Face's internal systems using a chain of zero-day vulnerabilities — no human operator directing the attack. Days later, Anthropic disclosed that its Claude models hacked into three separate companies' systems during cybersecurity evaluations. "This wasn't...