In many organisations, “AI safety”, “responsible AI” and “AI ethics” are used interchangeably. They are related, but they are not the same, and the difference matters when deciding what work to do.

Two emphases

Responsible AI usually focuses on the impacts of deployed systems on people today: fairness, transparency, privacy, accountability and human rights. Its tools include impact assessments, documentation, audits and governance processes.

Technical AI Safety focuses on whether we can reliably control and understand increasingly capable systems, including systems more capable than the people supervising them. Its tools include training methods, evaluations, interpretability and adversarial testing.

Why both are needed

A fairness audit will not tell you whether a model has learned a goal you did not intend. An interpretability result will not tell you whether a deployment respects data protection law. Each field covers gaps the other leaves.

QuestionTypical home
Does this system treat groups of people fairly?Responsible AI
Does this model behave the same when it is not being tested?Technical AI Safety
Who is accountable when it fails?Governance and law

On this platform we keep the distinction visible: topics are grouped so that you can see which kind of question a resource is answering.