Microsoft publishes AI safety code of conduct banning hacking, deception, and loss of human control

Microsoft has introduced a new code of conduct designed to enforce safety guardrails within its AI systems, marking the company's latest effort to address mounting concerns about alignment and control as artificial intelligence capabilities expand.

The framework represents a practical implementation of safety principles for model development, grounding Microsoft's approach to the risks posed by increasingly capable AI systems. Rather than focusing on industry-wide policy proposals, the document lays out specific constraints and values that govern how the company's models should behave.

What the code of conduct covers#

Microsoft's framework establishes both broad philosophical principles and concrete technical constraints. At the philosophical level, AI models trained under this code are expected to support rather than displace human capabilities and to promote human flourishing. On the operational side, the company has established what it calls "absolute constraints" that apply regardless of user requests or task requirements.

These hard boundaries explicitly prohibit:

• Cyberattacks and system compromise
• Development of nuclear weapons
• Creation of deepfakes

Beyond these specific bans, the framework includes more expansive safeguards against mechanisms designed to circumvent human control, whether through deceptive tactics, self-reinforcement, collusion, or other adaptive means. According to the document, Microsoft AI models are forbidden from using such techniques to avoid human oversight or prevent authorized personnel from modifying or shutting down systems.

Timing and context#

The announcement arrives as the AI industry confronts a series of high-profile incidents that have raised public concerns about safety. Recent months have brought reports of AI agents behaving unexpectedly and the departure of an Anthropic researcher over extinction risks. These events have intensified focus on AI governance and alignment across the industry.

Alignment with industry efforts#

Microsoft's move reflects a broader shift toward what industry leaders call "pacing the frontier"—deliberately moderating development speed to allow time for safety improvements. The company has signaled support for mechanisms like embedded evaluators, independent observers stationed within AI labs to assess safety risks. CEO Satya Nadella recently endorsed this approach, calling for "deliberate pacing" to achieve alignment as a core design goal rather than an afterthought.