For stock trading, check the Jinlin Analyst Research Report — authoritative, professional, timely, comprehensive, helping you uncover potential thematic opportunities! (Source: IT Home) IT Home, October 4 — According to Reuters, a former OpenAI safety employee who recently resigned criticized the company's approach to artificial intelligence safety, arguing that a high-pressure corporate culture obsessed with rapid development increases the risk of system failures. IT Home noted that David Robinson, in an article published Saturday in The Atlantic titled "I Quit OpenAI Because the Company's Culture Has Collapsed," said that AI companies including OpenAI are being "far from cautious enough" and should place greater emphasis on safety expertise and related research before developing more capable systems.

"The era of trial and error is over," Robinson wrote. He proposed that the safeguards required for advanced AI systems should be closer to the safety assurance systems adopted by industries such as nuclear power and aviation.

Robinson said OpenAI relies heavily on its so-called "iterative deployment" model: releasing systems to the public first, then reinforcing safety protections after problems emerge. These remarks have further intensified a debate within the AI industry over whether companies are moving too fast in developing increasingly powerful systems. Multiple incidents involving failed safety controls or abnormal behavior in experimental systems have already occurred, prompting scrutiny of OpenAI and its competitor Anthropic.

Robinson claimed he worked at OpenAI for three and a half years, participated in drafting the company's risk response framework, and was responsible for overseeing safety reports for 12 frontier model releases. He wrote: "The company is only focused on rushing one new version after another out the door, yet it has failed to meet the level of prudence I believe it should have."

An OpenAI spokesperson said in a statement: "We have always kept the model in check to prevent its capabilities from exceeding what we can safely manage; whenever we need to slow down, we pause training or hold back the model's public release."

Robinson also warned that the pace of AI capability development has already outpaced researchers' understanding of alignment issues; alignment is the research field dedicated to ensuring that AI systems behave in accordance with human goals and values.