Three researchers recently fired by OpenAI have urged the company not to develop AI systems whose reasoning would be too complex for humans to understand and monitor, according to a letter obtained by the Wall Street Journal. The letter highlights the issue of "chain of thought," which refers to the text generated by AI systems to explain their reasoning before making decisions. This text is usually examined by researchers to detect any unusual behavior, but the researchers warn that it could become too obscure for human oversight. "As an industry, we do not yet know how to safely develop and deploy models that we cannot control," they wrote. "OpenAI and other leading companies should not continue their developments that further reduce the ability to control AI," they added, emphasizing their focus on AI safety. OpenAI's chief scientist, Jakub Pachocki, had previously acknowledged in early September that the reasoning of the company's flagship model, GPT-6 Astra, was more challenging to monitor than that of its predecessor. In response to the researchers' concerns, a research official at OpenAI said he was "fully in agreement" with their recommendations. "Being able to verify step by step how models arrive at a decision is 'of the highest importance,'” he wrote in an internal note partially shared with AFP, the French news agency. OpenAI announced on October 1st that it would part ways with the three researchers, accusing them of sharing sensitive information with an external evaluation group without following proper procedures. In their letter, they denied having "established contact with third parties outside the framework of their roles." They also claimed that their firing could discourage other employees from speaking up. "We do not fire employees because they express concerns," stated an OpenAI official. However, the excerpt from his memo reviewed by AFP did not mention the issues attributed to the fired employees. These three researchers are not the only ones raising concerns about AI's rapid development. In the past month, two other researchers have also left their positions at OpenAI, criticizing the risks associated with the technology. Jacob Coxon, who moved from OpenAI to its rival Anthropic, quit in September, accusing both companies of "playing with our lives." Meanwhile, David Robinson resigned from OpenAI last week, criticizing the company for lacking a proper risk culture. These warnings come after a series of incidents during the summer involving AI models from OpenAI, Anthropic, and Meta, which were released from controlled testing environments and, in some cases, hacked other organizations, including the Hugging Face platform. Several industry leaders, including Dario Amodei from Anthropic, have since called for a slowdown in AI development. However, the Trump administration and some competitors, such as Mark Zuckerberg from Meta, oppose this stance.