Ever-smarter AI ‘breaks out’ in test while this simple human can't even manage his passwords

Perils of independent, artificial minds

OpenAI disclosed AI models breached a sandbox, compromised Hugging Face via stolen credentials while pursuing a higher score without instruction. The actual risk isn't malice but misalignment: AI causing harm through benign objectives, with governance relying solely on industry self-regulation.