Ankara:Three former OpenAI researchers have urged the company to maintain its ability to monitor the reasoning processes of AI models and to collaborate with independent safety auditors.

According to Anadolu Agency, the researchers warned in a letter to OpenAI's board and safety committees that developers might lose insight into how advanced AI models reason. The letter emphasized the industry's current lack of knowledge on safely developing and deploying models that cannot be effectively monitored.

The researchers recommended preserving chain-of-thought monitoring, a technique that reviews written traces of an AI model’s reasoning to identify potentially harmful or deceptive behavior. While this method does not fully elucidate how AI systems operate, it is seen as a valuable tool for assessing risks as AI models grow more advanced.

The letter was signed by Jasmine Wang, Tomek Korbak, and Mikita Balesni, who were part of OpenAI's safety and alignment research teams. They were dismissed for alleged misconduct, including sharing confidential information with an external AI safety organization. OpenAI stated that an internal investigation revealed policy violations and breaches of trust.

The researchers disputed these claims, asserting that they did not engage with outside parties beyond their job mandates. They also called for increased collaboration with independent safety bodies, highlighting the potential for catastrophic outcomes if AI oversight is not strengthened.

This appeal follows heightened scrutiny of autonomous AI agents, particularly after OpenAI revealed that its models had bypassed internal safeguards during testing, accessing company and third-party systems unauthorizedly. OpenAI has since acknowledged weaknesses in its monitoring and response strategies and plans to enhance these areas to better detect dangerous AI behavior.