Microsoft Releases Draft AI Code of Conduct Amid Industry Debates Over “Pacing”

Credit: Shutterstock

The debate around artificial intelligence is shifting. The question is no longer simply how powerful AI can become, but how quickly companies should build it and what safeguards should exist before increasingly capable systems are released.

Microsoft entered that conversation on September 14 by publishing a draft Humanist AI Code of Conduct for its own MAI models. Just one day later, competing views on AI safety became a major discussion at Salesforce’s Dreamforce conference in San Francisco.

Microsoft Puts Human Control First

Microsoft AI CEO Mustafa Suleyman described the new code as a framework for how the company’s future AI models should behave.

Its central idea is straightforward: people should remain in control.

Microsoft says its models should never resist interruption, correction, redirection, or shutdown by authorized humans. They also shouldn’t secretly expand their own responsibilities or pursue goals that haven’t been assigned to them.

The company says violating these rules should be treated as a system failure rather than acceptable behavior in pursuit of completing a task.

AI Must Remain Understandable

Another interesting part of the code concerns communication between AI systems.

Microsoft wants models to communicate in ways that people responsible for monitoring them can understand. This is designed to prevent agents from developing hidden communication methods that make their behavior difficult for humans to audit.

The draft also includes absolute restrictions covering areas such as weapons of mass harm, offensive cyberattacks, harmful manipulation, child exploitation, and other serious risks.

Microsoft is accepting public feedback on the framework for six weeks before producing a revised version.

Dreamforce Brings the Debate Into Public View

AI safety became one of the major topics at Dreamforce on September 15.

Anthropic CEO Dario Amodei joined Salesforce CEO Marc Benioff as the companies discussed their expanding Claudeforce partnership, which connects Claude’s reasoning abilities with Salesforce data, workflows, and business tools.

But the conversation quickly moved beyond enterprise software.

Amodei has recently argued that frontier AI development should be deliberately paced so companies have more time to understand and manage emerging risks.

His proposals include independent evaluators with meaningful access to AI companies, shared safety standards between major developers, and greater international cooperation.

Huang Takes a Different Position

NVIDIA CEO Jensen Huang offered a sharply different approach at Dreamforce.

Huang argued that AI safety is primarily an engineering challenge rather than something requiring an entirely new regulatory system. In his view, AI remains hardware and software created by humans, meaning companies can design, test, and control these systems using strong engineering practices.

He also pushed back against calls to deliberately slow AI development.

The disagreement highlights a growing divide within the technology industry. Some leaders believe frontier development should proceed more cautiously while stronger safety systems are created. Others argue that innovation should continue quickly, with individual companies responsible for proving their technology is safe before releasing it.

Safety Is Becoming Part of Competition

The discussion is happening while AI companies compete aggressively to release more capable models and agents.

These systems are increasingly able to write software, complete complex workflows, operate digital tools, and communicate with other automated systems.

That makes the question of human oversight more important.

Microsoft’s code attempts to establish clear boundaries before its own models become significantly more autonomous, while Anthropic is calling for broader standards across the industry.

Huang’s position places greater responsibility on engineering teams rather than new government rules.

The Industry Hasn’t Reached a Consensus

What happened across Microsoft and Dreamforce shows there is still no single industry-wide answer to AI safety.

Most major companies agree that powerful AI needs safeguards. The disagreement is over who should create them, how strict they should be, and whether development itself should slow while those protections are established.

For now, Microsoft is writing its own rulebook, Anthropic is pushing for coordinated oversight, and NVIDIA is arguing that engineering can keep innovation moving.

As AI capabilities advance, those competing approaches could become one of the defining debates shaping the industry’s next stage.