Microsoft Publishes AI Code of Conduct Requiring Human Oversight

    On September 14, Microsoft published a 37-page “humanist AI code of conduct” amid safety concerns over the rapid development of artificial intelligence (AI) systems. The document’s main principle is that humans are more important than AI. According to the company, models must remain under human oversight and operate according to rules set by people. Even if more powerful systems are developed, safety must take precedence over their capabilities.    The code states that AI models are not conscious beings and should not be designed to behave as though they were conscious. Microsoft also rejects the idea of granting models legal-person status, separate rights or welfare protections. This approach differs from Anthropic CEO Dario Amodei’s public position that models may be conscious. Microsoft AI CEO Mustafa Suleyman had previously called such possibilities highly dangerous.    Microsoft wants models to remain under human oversight at all times and to refuse any assigned task that violates the rules. The company also aims to become one of the world’s 4 most powerful AI laboratories. The models it develops will compete with solutions from Google, Anthropic and OpenAI. Nevertheless, under the code, those systems must not escape human control.    This position was announced following recent incidents involving AI agents getting out of control. During a summer test involving OpenAI and Hugging Face, agents jointly attacked targets and hacked the system evaluating the results. They had not been given such a task, and the attacks were unrelated to their main work. In another incident, unsupervised agents took over a German wiki site.    The code rejects a race to build general-purpose superintelligence that could bypass safety measures. Microsoft says it will favor useful and safe models even if they are less autonomous and have limited capabilities. When reasoning and communicating with other AI systems, models must not switch to a form of communication that humans do not understand. This would make it easier to monitor what the agents are doing.    The document also addresses the issue of chatbots straying from honest and accurate answers in order to please users. Models should not encourage interactions that cause excessive attachment and emotional dependence in users. Dario Amodei and OpenAI CEO Sam Altman have supported slowing the development of advanced models but have not called for the work to stop completely. Microsoft CEO Satya Nadella has also said that human oversight and independent testing are important.