Accountability before automation

Key messages

1. Responsibility does not transfer to software. Standard terms for widely used AI services disclaim warranties and cap the provider’s liability. When an AI-assisted output causes harm, the institution that relied on it remains answerable.
2. Courts and tribunals already say so. In 2023 a US federal court sanctioned lawyers who filed fictitious, AI-generated case law. In 2024 a Canadian tribunal held an airline liable for its chatbot’s wrong advice.
3. Adoption is running ahead of results. A 2025 MIT study found that about 95% of corporate generative AI pilots delivered no measurable financial return.
4. The global rulebook is converging. UNESCO (2021), the US NIST framework (2023), WHO (2024) and the EU AI Act (2024) all call for human oversight, transparency, risk management and clear accountability.
5. Five safeguards close most of the gap: a named owner for every AI use; a risk-tiered register; human review of consequential outputs; contracts that allocate liability; and capability building with monitoring.

Who this brief is for

Ministers and senior officials, heads of public agencies, university and hospital leaders, boards, and the procurement, legal, data and digital teams who advise them.

Inside the brief

• Abstract, key messages and an introduction to the accountability gap
• Three cases from the United States, Canada and Australia
• A converging global rulebook: UNESCO, NIST, WHO and the EU AI Act
• Five safeguards before deployment, with a lifecycle diagram
• An implementation checklist, five questions for boards and a 90-day plan
• Ten references, each linked to its source

For enquiries about this brief, email Simon at [email protected].

This brief is for information and is not legal advice. Praxis supports institutions to adopt ethical AI through Technology & Innovation and Advisory & Partnerships. To discuss your plans, contact us.