OpenAI safety employee quits, says ’time for trial and error is over’

A former OpenAI safety employee has resigned, claiming the company's fast-paced culture jeopardizes AI safety. In an article for The Atlantic, David Robinson argues that the pressure to rapidly develop AI systems increases the risk of failures. Robinson contends that advanced AI requires robust safeguards, similar to those in nuclear power and aviation industries.

He states that OpenAI relies on "iterative deployment," releasing systems and reinforcing safeguards only when issues arise. Robinson's comments fuel discussions about companies rushing to create increasingly powerful AI without adequate safety measures. OpenAI and rival Anthropic have faced criticism for safety control failures and unexpected system behavior.

Robinson, who worked at OpenAI for 3.5 years, drafted the company's preparedness framework and led safety reports for 12 frontier-model launches. He asserts that the company is failing to implement the necessary care, as they prioritize quick launches over thorough safety checks. An OpenAI spokesperson refutes Robinson's claims, stating that the company ensures models don't exceed safe management capabilities and pauses training or holds back models when needed.

Robinson also warns that AI capabilities are advancing faster than researchers' understanding of alignment, the discipline focused on making AI systems act according to human goals and values.