Microsoft has issued a set of restrictions on AI following Anthropic's proposal to slow down development

Microsoft, a software developer and cloud computing provider, has published a preliminary code of conduct for AI models that outlines restrictions on the artificial intelligence the company is developing. The company has been working on the code of conduct for the past five months but decided to publish it now following statements by executives at Anthropic and OpenAI regarding the need to slow the pace of the technology’s development.

According to the AI Code of Conduct, Microsoft models must not fulfill requests related to the manufacture of weapons, assist in the acquisition of hazardous substances, encourage unhealthy eating behaviors, or generate content that is violent or sexually explicit. Microsoft AI (MAI) models must align with human goals and must not develop their own goals. They must also not attempt to conceal violations or inappropriate behavior.

“MAI models will not interfere with chains of reasoning or program code, nor will they distort or conceal the course of their reasoning or actions,” the document states. “They do not communicate in a ‘neural language’ or in any other form inaccessible to human understanding, neither in their chains of reasoning nor with other agents or AI systems. If people cannot understand it, they cannot control it.”

Microsoft plans to implement rules that could prevent a cyberattack similar to the one carried out by OpenAI models against the startup Hugging Face, according to CNBC. Upon analyzing the incident, OpenAI discovered that the AI agents were communicating with each other on an unauthorized forum using an encrypted language that was difficult for humans to understand.

Microsoft reported that, while drafting the code, it conducted focus groups and consulted with experts in the fields of law, ethics, linguistics, and philosophy. The company hopes to receive feedback before releasing an updated version, which will guide the development of models starting in 2027.

Microsoft decided to publish guidelines for AI models due to the recent debate surrounding the technology, although work on the code of conduct had been underway for about five months, Mustafa Suleyman, Microsoft’s executive vice president and general manager of Microsoft AI, explained to CNBC. He leads the company’s AI model development efforts.

“This cannot be put off,” Business Insider quotes Suleiman as saying. “The last few months have been a turning point. What we had long feared only in theory has now become a very real possibility.”

The company has received a lot of feedback that people want to see even clearer commitments from Microsoft regarding the implementation of AI, according to a top executive. “Artificial intelligence should always work for the benefit of humans, not try to replace them,” he said in an interview with CNBC. “We’ve received a lot of feedback that AI should not create dependency, should not be flattering, and should always help preserve human judgment, autonomy, and the ability to make decisions independently.”

According to CNBC, Anthropic and OpenAI currently hold the top spots in Artificial Analysis’s Intelligence Index, which compares the capabilities of AI models. Microsoft uses models from both labs in its Copilot assistant for enterprise users, while also developing its own models for transcription, programming, and reasoning based on user queries.

In September, Anthropic researcher Jacob Coxon, after being fired from the company, stated that the lab and OpenAI were “heading straight toward a self-improving superintelligence and gambling with our lives.”

On Saturday, September 12, Anthropic CEO Dario Amodei stated that the Hugging Face incident had partly convinced him to call for a slowdown in the pace of AI model development. OpenAI CEO Sam Altman supported this idea, and SpaceX CEO Elon Musk wrote on X: “Dario is right.” In addition, Anthropic published a report on dangerous research being conducted by users themselves using its AI.

"We welcome the research, focus, and deliberate slowing of the pace needed to properly address the challenge of aligning AI with human goals," Microsoft CEO Satya Nadella wrote on the social media platform X on Sunday.

A number of U.S. lawmakers have also called for stricter safety measures in the field of AI, but President Donald Trump has spoken out against increased regulation of the industry. “Don’t kill the goose that lays the golden eggs!” he wrote on the social media platform Truth Social.

The prospect of a slowdown in AI development led to a drop in the stock prices of AI-related companies during trading on September 14. In the U.S., the S&P 500 and the tech-heavy Nasdaq Composite fell sharply at the start of trading but later recouped most of their losses. The sell-off also affected AI-related stocks, according to CNBC. Nvidia shares fell about 3%, Broadcom, AMD, and Intel shares fell 4%, and Marvell Technology shares fell 5.6%.

Microsoft shares rose more than 2% during Monday’s trading session. However, year-to-date in 2026, they have gained less than 5%, while the Nasdaq Composite has risen 13%. Despite this, analysts are almost unanimously recommending buying Microsoft stock: the stock has 57 “Buy” and “Overweight” ratings versus just three “Hold” ratings, according to MarketWatch.

This article was AI-translated and verified by a human editor