When building software with AI-assisted tools like GitHub Copilot CLI, it is important to follow Responsible AI practices. This document summarizes Microsoft's Responsible AI principles and provides actionable guidance for developers.
Microsoft's Responsible AI Standard is built on six core principles. These apply whether you are training models, consuming AI services, or using AI-powered developer tools.
AI systems should treat all people fairly and avoid affecting similarly situated groups in different ways. When using AI-generated code, review outputs for assumptions or biases — for example, hard-coded locale assumptions, gender-biased variable names, or datasets that aren't representative.
In practice:
- Review AI-generated code for cultural or demographic assumptions.
- Test with diverse inputs and edge cases.
- Use inclusive language in prompts, code, and documentation.
AI systems should perform reliably and safely under both expected and unexpected conditions. AI-generated code must be tested and validated just like human-written code — it is not inherently correct.
In practice:
- Never ship AI-generated code without review and testing.
- Validate outputs against your specification and acceptance criteria.
- Treat AI suggestions as a starting point, not a final answer.
AI systems should respect privacy and protect personal and business information. Be mindful of what data you include in prompts or share with AI-powered tools.
In practice:
- Do not paste secrets, credentials, or personal data into AI prompts.
- Understand your tool's data handling policies (e.g., GitHub Copilot privacy).
- Follow your organization's data classification and handling policies.
AI systems should empower everyone and engage people of all abilities. When AI helps generate user-facing content or interfaces, ensure the results are accessible and inclusive.
In practice:
- Ensure AI-generated UI code includes proper accessibility attributes (ARIA labels, semantic HTML, keyboard navigation).
- Consider localization and internationalization from the start.
- Design for diverse users, including those with disabilities.
AI systems should be understandable. Users and stakeholders should know when AI has been used and understand its limitations.
In practice:
- Document when and how AI tools were used in your development process.
- Be transparent with your team about AI-assisted contributions.
- Understand that AI models can hallucinate — outputs may look correct but be factually wrong.
People are accountable for AI systems they design, deploy, and use. The developer — not the AI tool — is responsible for the final code.
In practice:
- You own the code you commit, regardless of how it was generated.
- Establish code review processes that account for AI-generated code.
- Maintain human oversight and decision-making authority.
This workshop uses GitHub Copilot CLI as an AI-assisted development tool. Here is how the principles apply:
| Principle | Workshop Application |
|---|---|
| Fairness | Review generated code for biased assumptions |
| Reliability & Safety | Validate all outputs against the SDD specification |
| Privacy & Security | Avoid sharing sensitive data in prompts |
| Inclusiveness | Ensure generated code meets accessibility standards |
| Transparency | Acknowledge AI assistance in your workflow |
| Accountability | You are responsible for every line you commit |
- AI is a tool, not an authority. Always review, test, and validate AI-generated outputs.
- Keep humans in the loop. Maintain meaningful human oversight at every stage.
- Protect sensitive data. Be deliberate about what you share with AI systems.
- Build inclusively. Use AI to help reach more people, not fewer.
- Stay accountable. The developer is always responsible for the final product.