Trust at Scale: Why Governance Can't Be an Afterthought in Enterprise AI
Karthik Mani

Most organisations learn the same lesson too late: trust cannot be added to a system after it is built. A capability ships under pressure, works, and spreads — then an audit, an incident, or a regulator asks who approved this, what data it used, why it decided that, and the organisation discovers the information was never captured and the accountability never assigned. What follows is a retrofit costing several times what the original controls would have. Artificial intelligence has raised the stakes of this old lesson, because systems that decide and act autonomously are less able to answer those questions by default and the cost of being unable to answer is larger. Drawing on two decades designing enterprise security, access-governance, and AI-enabled systems across three countries, this book makes one argument in many forms: trustworthy AI is an architectural discipline, not a compliance checkbox. It builds the trust stack layer by layer — identity, data, access, control, accountability, assurance — then treats each as an engineering problem with concrete patterns: policy enforcement points, immutable records, fail-closed defaults, bounded blast radius, auditable decisions, bounded autonomy, and data protection across its full lifecycle. Later chapters address what determines whether governance survives contact with a real organisation: the operating model, incident response and kill switches, sector case studies, third-party risk, regulatory alignment, and the roadmap from reactive controls to continuous assurance. For architects, engineering leaders, risk professionals, and executives who want AI systems trustworthy by design rather than by remediation.

Trust at Scale : Why Governance Can’t Be an Afterthought in Enterprise AI

Comentarios