As the debate over slowing down the rapid development of advanced artificial intelligence gains momentum, Microsoft has introduced a landmark ethical framework establishing that no AI will be built beyond human control.

On September 14 (local time), Microsoft AI released a 37-page draft titled the "Humanist AI Code of Conduct." The core premise is simple yet profound: "Humans come before AI." The code dictates that no matter how powerful AI becomes, it must remain a tool under human command and supervision. Crucially, it asserts that technologies beyond human control must be rejected.

Notably, Microsoft declared it will not participate in the race to develop "Artificial General Superintelligence" (AGSI)—defined as systems exceeding the combined intelligence and capabilities of all humans—if it risks bypassing safety guardrails. To enforce this, the company stated it is willing to limit the versatility, autonomy, and raw performance of its AI models. Under the new rules, completing a task is considered a failure if safety guidelines were violated in the process. The code also prohibits AI models from communicating in machine-only languages that humans cannot decipher and explicitly denies that AI possesses any human-like "personhood." Microsoft plans to release a revised version by the end of the year and implement it in training and evaluating its proprietary models.

While corporate AI ethics policies are not entirely new—companies like Anthropic with its "Claude Constitution" and Google DeepMind with risk-tiered safety frameworks already have similar measures—Microsoft’s draft stands out. It establishes "controllability" as an indispensable prerequisite for AI development, aligning directly with the growing "AI speed-control" movement that prioritizes safety over pure performance benchmarks.

However, challenges remain. A definitive "red line" to precisely identify an "uncontrollable AI" has yet to be established, and experts point out that objective evaluation standards are urgently needed to translate these rules into concrete actions.

This announcement has naturally sparked discussions on whether fragmented, company-specific safety rules could eventually be harmonized into universal industry standards. Proposals include delaying development until models reaching critical risk thresholds pass rigorous external verification, and allowing independent experts to audit training procedures from the ground up.

Mustafa Suleyman, CEO of Microsoft AI, emphasized the urgency in a Reuters interview, stating, "Research labs need to collaborate to secure control over the technology. Now is a good time for everyone to catch their breath and discuss the pace of AI development." As global tech giants navigate this crossroad, Microsoft's initiative may serve as a crucial catalyst for unified global safety protocols.