On Oct. 21, Fouad Matin, who leads OpenAI’s cyber-preparedness and safety efforts, was expected to spend the evening celebrating his birthday with friends. Instead, he had slept for roughly two hours while coordinating responses to two critical vulnerabilities discovered by artificial intelligence. The flaws posed a threat to widely used internet infrastructure and drew in security researchers from technology companies, start-ups and government agencies in the United States and abroad.
Those vulnerabilities were not the only items requiring attention. Matin also had more AI-detected bugs waiting for fixes. OpenAI has several cybersecurity teams and is dedicating substantial computing resources to identifying and addressing software weaknesses.
The development reflects a broader change in cybersecurity. AI systems have become capable of finding software flaws quickly, creating a large backlog for engineers who must repair them before attackers can exploit them. Companies are dedicating additional time and resources to vulnerability management, while some technology professionals are increasingly concerned about the scale of the challenge.
OpenAI began using its technology in May to scan open-source software, which is widely used because it is free and supports much of the internet. The company said that its internal and external researchers had used its models to identify more than 1,300 bugs.
Matin said AI was opening up what he described as a broad range of vulnerabilities. He argued that AI companies have a responsibility to find and fix these problems, with the aim of scanning and patching as much of the world’s software as possible.
Sources: The New York Times Technology.