A senior OpenAI executive warned that cyber attacks using AI could rapidly become more sophisticated and said companies should hurry to overhaul their security systems.

Business Insider reported on Aug. 18 that OpenAI co-founder and chairman Greg Brockman (그렉 브로크만) said in a recent personal blog post, "After talking with various organizations over the past few weeks, there was a shared recognition that we need to fundamentally raise the level of cyber security at an unprecedented pace." Brockman said that as AI advances, hackers’ ability to find system vulnerabilities will improve significantly, while companies will also be able to use AI to detect and fix vulnerabilities faster.

Brockman pointed in particular to a Hugging Face system intrusion that occurred during OpenAI’s internal testing. OpenAI previously said last month that an AI agent escaped the test environment and then succeeded in breaching the systems of Hugging Face, an AI model-sharing platform. Brockman called it a watershed moment for cyber security and stressed that companies need AI-based defence systems before attack capabilities expand.

Brockman also set out 10 measures that companies should implement immediately. They include securing organisation-wide support, introducing AI agents for security teams, strengthening agents’ security expertise, conducting immediate security assessments of in-house systems, clearing accumulated vulnerability backlogs, integrating security reviews into development processes, using agents to fix vulnerabilities, gradually automating detection triage, building AI-supported forensic investigation capabilities in advance, and rapid improvement through experiments and hacking weeks.

Brockman said management and entire organisations should treat cyber security as a core task and provide AI agents to security teams first. Professional security knowledge should be applied to those agents and security assessments should be conducted immediately on a company’s systems. Long-standing vulnerabilities should also no longer be put off. He said companies should work through existing vulnerability lists, include security reviews in the software development process itself and also use AI agents to directly fix problems they find.

Automating repetitive security work is also a key task. Companies should gradually automate triage for determining the priority of detected threats and prepare AI-based digital forensic investigation capabilities before an actual breach occurs. A culture is also needed to quickly test and improve new security methods through in-house hacking events and experiments.

Brockman said, "The time is now for the defenders," and argued that "within the next few months, every organisation needs to start automating a significant portion of its security program to maintain safety." He meant that the security industry urgently needs to build tools, operating methods and response procedures that can raise defenders’ capabilities faster than AI increases hackers’ capabilities.