OpenAI proposes stronger AI safeguards and more support for cyber defenders

OpenAI published a policy blueprint on October 21 urging governments and AI developers to strengthen cybersecurity defenses. The company said defenders have limited time to prepare before powerful AI capabilities become more widely available to attackers.

The proposals cover model security, cybersecurity modernization and government preparedness. They call for clearer responsibility for securing advanced systems, detecting harmful behavior and maintaining effective shutdown mechanisms. OpenAI also wants defenders to receive better tools, training and technical assistance.

The company said security incidents and unintended behavior involving its own models helped shape the recommendations. The document builds on earlier policy proposals and argues that AI can help identify vulnerabilities and respond to intrusions while also creating opportunities for misuse.

OpenAI urged policymakers to strengthen government testing, independent assessments and incident reporting. It also called for investment in aging systems and essential services, with requirements adjusted as AI capabilities develop.