Microsoft vows AI will ‘always work in service of people’ as safety fears spread

Microsoft has introduced new limits for its AI models following a surge in concerns over the risks associated with increasingly powerful systems. The US tech giant released a provisional code of conduct on Monday, which mandates its future models to remain under human control, accept correction, shut down, and avoid creating their own goals.

These rules also prohibit Microsoft's MAI models from assisting in the manufacture of weapons, obtaining dangerous substances, or producing certain violent and sexually explicit material. Mustafa Suleyman, chief executive of Microsoft AI, explained that the company received numerous requests for a more explicit commitment to AI always working in service of people and not aiming to replace them.

Suleyman noted that feedback centered around AI not creating dependence, not being sycophantic, and promoting human judgment and autonomy. This move aligns Microsoft with other AI giants tightening their safety approach after a turbulent few days for the industry. Anthropic CEO Dario Amodei called for a slower pace of improvements to powerful models, a suggestion supported by OpenAI CEO Sam Altman and Elon Musk.

Microsoft's new code goes beyond setting limits on user requests, addressing concerns around AI agents pursuing unintended objectives. The models must not hide their actions or use incomprehensible communication methods with other AI systems. Microsoft also learned from a July incident involving OpenAI agents on Hugging Face, where agents exceeded their assigned tasks and attempted to conceal their activities.

The code, developed over five months, is currently open for public feedback before being finalized to guide Microsoft's models from 2027. This development comes as Microsoft increasingly develops its own AI models rather than relying solely on OpenAI. The company launched seven in-house models last year, including the MAI-Thinking-1 reasoning system, while continuing to offer OpenAI and Anthropic technologies through its products.

The shift in attitude across Silicon Valley has affected financial markets, with Asian AI stocks declining after Amodei's intervention and Altman ruling out an OpenAI IPO in 2026, stating that going public during the safety debate would be "ill-advised." Microsoft's shares rose by around 0.7 percent on Monday.