The principles landscape
Governments, international bodies and AI companies have all published responsible AI principles. Here's how they compare — and what they share.
Learning objectives
- Summarise the OECD, UNESCO and UK principles
- Compare major AI companies' responsible AI frameworks
- Identify six core principles common to most frameworks
- Explain 'ethics washing' and what closes the principles-practice gap
2.1 International frameworks
Hundreds of AI ethics principles have been published. A few are especially influential:
| Framework | Who | Core principles |
|---|---|---|
| OECD AI Principles (2019, updated May 2024) | Adopted by OECD members and other countries; basis for the G20 AI principles | Inclusive growth and well-being; human rights and democratic values including fairness and privacy; transparency and explainability; robustness, security and safety; accountability |
| UNESCO Recommendation on the Ethics of AI (2021) | Adopted by UNESCO's member states | Human rights and dignity, environmental flourishing, diversity and inclusion, peace and justice — with policy action areas |
| UK AI regulation principles (2023 white paper) | UK Government, applied by existing regulators | Safety, security and robustness; appropriate transparency and explainability; fairness; accountability and governance; contestability and redress |
2.2 What the major AI companies commit to
| Organisation | Framework |
|---|---|
| Microsoft | Six principles: fairness; reliability and safety; privacy and security; inclusiveness; transparency; accountability — put into practice through its Responsible AI Standard and annual transparency reports. |
| AI Principles organised around bold innovation, responsible development and deployment, and collaborative progress — including human oversight, safety research, bias mitigation and privacy. | |
| AWS | Eight dimensions: fairness; explainability; privacy and security; safety; controllability; veracity and robustness; governance; transparency. |
| IBM | Principles of trust and transparency, with pillars including explainability, fairness, robustness, transparency and privacy. |
| Anthropic | Publishes Claude's Constitution — the values and priorities its models are trained to follow — and a Responsible Scaling Policy setting safety measures for more capable models. |
| OpenAI | Publishes usage policies for what its models may be used for, plus safety frameworks and system cards for its models. |
2.3 The common core
Despite different wording, almost every framework converges on the same handful of ideas. This course uses six:
Fairness
AI should not create or reinforce unjust discrimination. (Module 3)
Transparency & explainability
People should know when AI is used and be able to understand relevant decisions. (Module 4)
Accountability
Identifiable people and organisations answer for AI systems and their effects. (Module 4)
Privacy
Personal data is protected and used lawfully and proportionately. (Module 5)
Safety & robustness
Systems work reliably, resist misuse and fail safely. (Module 5)
Human agency & oversight
People stay in meaningful control, and AI supports rather than undermines autonomy. (Module 5)
2.4 From principles to practice
Principles are easy to publish and hard to live by. Critics call the gap "ethics washing": impressive values with no change in practice. What closes the gap:
- Concrete requirements derived from each principle (e.g. "test performance across demographic groups before launch")
- Roles with responsibility and authority
- Review points in the product lifecycle where work can be paused or stopped
- Tools for testing, documentation and monitoring
- Transparency about what was done — and what went wrong
Pick two organisations' principles from the table (or your own organisation's, if it has some). Which of the six core ideas does each cover? What's missing? What evidence would convince you they're applied in practice?
Key takeaways
- The OECD Principles (updated 2024) and UNESCO Recommendation are global reference points.
- The UK applies five cross-sector principles through existing regulators.
- Company frameworks differ in wording but converge on a common core.
- Principles only matter with requirements, roles, review points, tools and transparency.