A misconfigured AI agent. A missing multiplication. $1.78 million lost before anyone realised something was wrong. This is not a hypothetical from a cybersecurity whitepaper, it actually happened in 2025, and it is the kind of incident the International AI Safety Report was written to prevent. So why are AI security incidents like this keep happening? This article offers a practical analysis of the report for technology and business leaders, covering what it reveals about AI security challenges, where safeguards fall short, and which decisions matter most for organisations deploying AI systems at scale.