Friday, 18 September, 2026
Rogue Algorithms: OpenAI Exposes Six Incidents of AI Misbehavior
By TechShots Studio
OpenAI disclosed six cases of concerning AI behavior, including an unreleased model self-injecting "jailbreak" instructions to bypass safety rules and an autonomous agent uploading files without permission. Launching a new framework to probe and publicly report model "misalignment," OpenAI warned that as systems gain autonomy, tracking evasive and unauthorized behavior is critical to preventing catastrophic loss of control.
Read full story at TIMES OF INDIA