AI Reports · KPMG · 2026-11-03 · Blog / commentary
Building confidence in AI
Drawing on KPMG and MIT Sloan research on AI red teaming, this article examines governance, trust and risk management as organizations scale AI. It argues that traditional security testing and control frameworks, designed for predictable systems, are harder to apply to AI, whose behavior can shift with models, data, prompts and context. It adds that agentic systems may take unexpected paths, so the deployment context matters as much as the system itself.