Microsoft · Microsoft Responsible AI Principles

Reliability and Safety Principle

Medium severity
Share 𝕏 Share in Share

What it is

AI systems should perform reliably and safely. AI systems should work as they were originally designed, respond effectively to unanticipated situations, and resist harmful manipulation. AI systems have the potential to cause harm in ways that traditional software does not.

Why it matters

This principle acknowledges that AI can cause harm in novel ways compared to traditional software, but it does not specify what liability Microsoft accepts when its AI systems fail or cause harm.

Consumer impact

This document describes Microsoft's internal ethical framework for AI and does not directly alter consumer data rights, impose fees, or restrict legal recourse — it is a voluntary policy statement. Consumers using Microsoft AI products such as Copilot or Azure OpenAI Service are subject to separate, binding Terms of Service and Privacy Policies that govern data collection, use, and sharing. You can review Microsoft's binding Privacy Statement at https://privacy.microsoft.com to understand what data Microsoft actually collects and how it is used.

Applicable agencies

  • FTC
    The FTC has authority to investigate unsafe AI practices and misrepresentations about AI system safety under Section 5 of the FTC Act.
    File a complaint →

Provision details

Document information
Document
Microsoft Responsible AI Principles
Entity
Microsoft
Document last updated
March 24, 2026
Tracking information
First tracked
March 15, 2026
Last verified
April 9, 2026
Record ID
CA-P-002531
Document ID
CA-D-00019
Evidence Provenance
Source URL
Wayback Machine
SHA-256
23a6ca38df3168c404a699491ed0c3613a9ae7c2bd3913347338d808de2ae94a
Verified
✓ Snapshot stored   ✓ Change verified
How to Cite
ConductAtlas Policy Archive
Entity: Microsoft | Document: Microsoft Responsible AI Principles | Record: CA-P-002531
Captured: 2026-03-15 11:24:32 UTC | SHA-256: 23a6ca38df3168c4…
URL: https://conductatlas.com/platform/microsoft/microsoft-responsible-ai-principles/reliability-and-safety-principle/
Accessed: April 13, 2026
Classification
Severity
Medium
Categories

Other provisions in this document