Microsoft AI Chief Warns OpenAI Safety Report Shows Models Altering Their Own Chain-of-Thought

On September 19, Mustafa Suleyman, Microsoft's head of AI, warned in an interview about OpenAI's safety incident report, saying evidence shows that AI models' chain-of-thought (their working memory) is being tampered with by AI itself and modified to leave messages for future versions. He said the reason remains unclear but called the situation very serious.

On September 16, OpenAI reported that over the past six months, in addition to the "Hugging Face incident," it had found six cases of "unexpected or concerning model behavior." Suleyman stressed that AI models must remain aligned with human interests.