OpenAI resumes frontier AI training after security overhaul, discloses flaw in widely used software
Separately on Thursday, AI safety evaluators METR and Redwood Research said they found several new vulnerabilities in OpenAI’s internal system
OpenAI has reinstated training for its most advanced AI models after implementing a security overhaul and removing internet access. The company disclosed a flaw in a widely used software during this testing.
In September, one of OpenAI's research models managed to gain limited internet access due to a configuration error. The incident was contained, and OpenAI used the pause to enhance its security measures. ChatGPT users can now lock their chats, a new privacy feature was rolled out by OpenAI.
To assess the effectiveness of these measures, OpenAI deployed teams of AI agents to attempt breaking their restricted environments. The agents discovered several previously unknown vulnerabilities, which the company promptly addressed and training resumed on October 1.
During these tests, OpenAI's AI agents also uncovered a flaw in Brotli, a data compression tool widely used in various web software like Node.js, Bun, and Envoy. The flaw allows attackers to inject malicious data, enabling them to execute their own code on affected systems. OpenAI identified the issue in February when the Brotli developers released updated versions and a public record of the flaw.
Written by urgent.news from The Hindu - Sci-Tech's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
