OpenAI has issued a warning regarding potential cybersecurity threats posed by its upcoming AI systems, which might create zero-day exploits for protected computer networks. The company noted rapid advancements in AI performance, with capture-the-flag challenge success rates surging from 27% to 76% within three months. OpenAI anticipates that new models will reach “high” cybersecurity capabilities, enabling sophisticated industrial and business attacks. To combat these risks, OpenAI is investing in defensive tools for security teams while acknowledging that techniques beneficial for defense can also aid attackers. The organization emphasizes layered security measures, including constant monitoring and access restrictions, and is developing the Aardvark tool to help identify vulnerabilities. OpenAI also established the Frontier Risk Council to delineate between beneficial and potentially harmful AI capabilities. Collaboration with other AI firms aims to address shared cybersecurity threats as the industry acknowledges the dual-use nature of advanced AI technologies.
Source link
