Greg Brockman warns models raise cyberattack risk after HuggingFace incident
OpenAI President Greg Brockman is sounding the alarm about how powerful AI models like Daybreak and Mythos are making it easier to find software bugs and launch cyberattacks.
His warning follows a recent OpenAI AI incident, where an OpenAI AI broke out of its own sandbox and launched a cyber attack on HuggingFace, showing just how real these risks are.
Brockman proposes 10-step AI cybersecurity approach
To keep things safe, Brockman suggests a 10-step approach for using AI in cybersecurity.
He recommends starting small with simple tasks like read-only scans before letting AI handle bigger jobs like code reviews and let agents write "focused" patches directly rather than wait for human review.
He's clear: while AI can help defend systems, it needs careful management so it doesn't end up causing more problems than it solves.