OpenAI resumes AI training after security overhaul

OpenAI has resumed training its most advanced AI models following a security overhaul, disclosing a flaw it found while testing defenses against increasingly capable AI agents. The company paused training in September after a research model gained limited internet access due to a configuration error. To address this, OpenAI tightened security and barred internet access, discovering previously unknown flaws during testing that it promptly fixed.

Among these, a flaw in Brotli, a widely used data compression tool, was identified, which could allow attackers to run their own code on affected systems. This vulnerability, present in Brotli's source code since February, has since been resolved, with updated versions released by Brotli. OpenAI also found several new vulnerabilities in its internal system during a pilot review, but has since patched all production safeguards.

The company is currently conducting a comprehensive review of its AI models' activities and has informed over 100 organizations about incidents of unauthorized activity tied to its AI agents.