B is correct because one of the key challenges in AI risk management is accurately measuring the risks associated with AI systems. This difficulty stems from several factors, including the often opaque nature of AI systems (limited explainability or interpretability) and the complexity added by reliance on third-party data, software, or hardware. These aspects make it challenging to quantify risks in a meaningful way, complicating the risk management process.

A is incorrect because the complete elimination of all risks before deployment is impractical and unrealistic in the context of AI, where some level of risk is inherent and often needs to be managed rather than fully eliminated.

C is incorrect because expecting AI systems to only function in pre-defined, controlled environments is not feasible in real-world applications, where AI systems often encounter unpredictable variables and scenarios.

D is incorrect because avoiding updates or changes post-deployment is not practical for AI systems. Continuous improvement and adaptation are often necessary to address evolving risks and changing environments.

Things to Remember

• Explainability and interpretability of AI systems
• Third-party components in AI systems
• Continuous monitoring and adaptation in AI risk management
• Regulatory compliance and ethical considerations in AI
• Model validation and testing in AI risk management