The company says its new AI model can find unknown vulnerabilities without human guidance, and access will be restricted as a result.
