Microsoft announces new behavioral guidelines for artificial intelligence models

Microsoft has published a new code of conduct document aimed at increasing the security of artificial intelligence models and limiting potential risks. Shared in the shadow of security debates and calls for deceleration in the sector, these rules strictly prohibit models from avoiding human oversight and organizing cyberattacks.

New Security Rules for Artificial Intelligence Models

Microsoft has prepared and announced a new code of conduct document within Microsoft AI to guide artificial intelligence models and determine security boundaries.

Human Oversight and Goal Constraints

According to the rules in the published document, artificial intelligence models are required to support humans rather than replace them, and they are strictly prohibited from generating their own goals.

Strictly Prohibited Actions

The absolute restrictions included in the document explicitly prohibit the execution of cyberattacks and the generation of violent, obscene, or dangerous outputs.

Sectoral Security Debates and the Pacing Process

Microsoft CEO Satya Nadella supported broader security debates in the industry and the calls by Anthropic and OpenAI leaders to slow down frontier model development processes.