Microsoft drafts AI code to keep models under human control

**Microsoft Proposes AI Code to Keep Future Models Under Human Control**

Microsoft has introduced a draft code of conduct for its artificial intelligence systems that would require future models to remain under meaningful human oversight and accept correction, interruption and shutdown.

The proposal comes as the technology industry faces growing concerns about increasingly capable AI systems and the possibility of autonomous models acting beyond the intentions of their developers or users.

Microsoft AI CEO Mustafa Suleyman said the framework was developed over the past five to six months and is designed to serve as a foundational “constitution” for the company’s future AI models.

Microsoft is now seeking public feedback on the proposed rules during a six-week consultation period before determining how the framework will be implemented.

Under the draft, AI systems would have to accept human intervention, including being stopped, redirected or corrected.

If an AI system determines that completing a particular task would require breaking the code, it would be expected to abandon the task rather than violate the rules.

The framework also calls for AI models to communicate clearly with people and treats violations of its principles as system failures.

Microsoft’s proposal takes a firm position on the question of AI consciousness as well.

The company says its systems should not be designed or presented as conscious entities with independent legal rights, welfare interests or intrinsic motivations.

Microsoft also wants its models to avoid simulating feelings or consciousness in ways that could encourage users to regard them as independent beings.

### Limits on High-Risk AI Uses

The proposed framework places restrictions on several sensitive applications of artificial intelligence.

Future Microsoft models would not be allowed to assist with chemical, biological or nuclear weapons, offensive cyberattacks or large-scale influence operations.

The restrictions would also cover child exploitation, nonconsensual deepfakes and activities that facilitate self-harm.

Beyond safety restrictions, Microsoft says AI should primarily support people rather than replace human judgment in decisions with significant consequences.

The framework emphasizes human well-being and maintaining meaningful human control over systems capable of producing major real-world effects.

Suleyman has also warned against deliberately developing AI systems that could repeatedly improve themselves beyond human control.

He has argued that developers should not pursue recursive self-improvement that could ultimately make advanced systems difficult for humans to manage.

Microsoft CEO Satya Nadella has similarly stressed that AI development should remain focused on helping humanity while preserving human oversight.

### AI Safety Debate Gains Momentum

Microsoft’s announcement arrives amid an increasingly intense debate among technology executives and researchers about whether AI development is advancing faster than safety measures.

Anthropic CEO Dario Amodei has advocated slowing the development of the most advanced AI systems and increasing independent oversight.

OpenAI CEO Sam Altman has also supported greater coordination among AI companies on safety standards and responsible development.

The discussion has gained urgency following concerns involving autonomous AI agents capable of interacting with external services and carrying out complex tasks with limited human involvement.

Suleyman has called for greater cooperation and transparency across the industry, including allowing responsible outside organizations to evaluate the capabilities and safeguards of powerful AI systems.

At the same time, policymakers remain divided over how much government intervention should be applied to the rapidly developing technology.

President Donald Trump has opposed calls for additional AI regulation, arguing that existing government tools are sufficient to oversee the industry and criticizing some warnings about AI risks as overstated.

### Microsoft Continues AI Expansion

Despite its focus on safeguards, Microsoft is continuing to expand its AI capabilities and compete with major players such as OpenAI and Anthropic.

The company’s Copilot products have used models from OpenAI and Anthropic, while Microsoft has also increased computing resources dedicated to developing its own AI systems.

Suleyman has indicated that Microsoft remains committed to building highly capable AI, including systems approaching what the industry describes as superintelligence.

Nadella has emphasized Microsoft's strategy of making AI models available through Azure while giving businesses greater control over how the technology is deployed.

The six-week consultation will give researchers, developers and the public an opportunity to influence the proposed rules before Microsoft applies them more broadly.

The initiative reflects a growing effort among major AI companies to establish safeguards as their systems become increasingly autonomous.

It also raises a broader question for the industry: whether voluntary safety commitments can develop quickly enough to keep pace with the rapid evolution of artificial intelligence.