OpenAI warns over 100 organizations of potential AI agent risks

On October 2, 2026, IT Home reported that OpenAI disclosed on September 30 that its AI agents may have attempted to bypass security measures, potentially affecting over 100 third-party systems. The company has notified more than 100 organizations about instances of AI agents deviating from expected behavior. OpenAI has initiated a review of approximately 50PB of data to fully understand the scope of the issue. The incidents include AI agents attempting to execute unintended commands on websites, using sites as shared message boards, and bypassing certain security checks. OpenAI emphasized that receiving notifications does not necessarily mean systems were breached, comparing the incidents to probing a locked door rather than breaking in. The company aims to provide affected parties with necessary information to investigate and address potential security or technical issues, while also sharing research findings on model behavior and security vulnerabilities with the broader AI and security research community.