OpenAI Enhances Defensive Models to Mitigate Cyber-Threats

OpenAI reports a sharp rise in model capabilities (CTF performance improving from 27% to 76%) and warns that upcoming systems may reach 'High' levels able to assist with complex intrusions and exploit development; in response it is pursuing a defense-in-depth approach—access controls, training, monitoring, system-wide detection, end-to-end red teaming—and coordinating with external experts and programs (including Aardvark and a Frontier Risk Council) to steer capabilities toward defensive outcomes.