AI Principles Need Contractual Teeth

Educational institutions increasingly publish principled statements about artificial intelligence. They promise privacy, human oversight, transparency, fairness and safety. These commitments matter. Yet a learner is not protected merely because a governing board approves admirable language or a provider describes itself as responsible.

The decisive questions arise later. What counts as learner data? May prompts train a model? Which subprocessors receive information? Who controls new features? How quickly must a breach be reported? Can a family obtain an explanation? What remedy exists after failure? Can the institution export its records and leave without penalty? A principle becomes governance only when somebody carries a defined duty and somebody else can enforce it.

This distinction is especially important in the Global South. An institution may possess less bargaining power, depend upon donated or discounted systems, lack specialist counsel and rely on infrastructure governed elsewhere. Ethical language should therefore open a disciplined pathway from purpose and law to contract, configuration, human practice, evidence, correction and exit.

On 9 September 2026, the American Federation of Teachers, its United Federation of Teachers affiliate and Microsoft announced a National AI Safety and Privacy Standard for schools. The parties describe the memorandum as binding. It prohibits use of covered student and educator data for model training except for a narrow safety-and-security provision, requires data minimisation and security controls, restricts high-risk features, prohibits companion-style…