OpenAI Safety Worker Resigns, Warning of Errors That Cannot Be Undone

David Robinson, an employee working on artificial intelligence safety at OpenAI, has resigned, criticizing what he described as a breakdown in the company’s workplace culture.

Robinson’s departure was accompanied by a warning about the risks of relying on identifying mistakes after they happen and then attempting to fix them.

His criticism centered on what he viewed as a failure to take risks seriously, raising concerns about whether a reactive approach is adequate for AI safety.

The warning drew a distinction between problems that can be corrected later and errors whose consequences cannot be undone. His concerns focused on the limits of treating fixes after the fact as a safeguard.