liveBook · Manning

Probabilistic risk management

As explained in AI Governance

Probabilistic risk management evaluates and governs risks whose likelihoods or outcomes are uncertain rather than deterministic.

A generative AI model may leak sensitive information even when conventional user permissions are respected. Probabilistic risk management addresses such uncertain outcomes without assuming that controls will behave deterministically. section 1.3 section 1.3

Put the principle into operation by combining established governance with adaptive practices:

  • Retain policies, accountability chains, executive oversight, and existing IT and API controls. section 1.3
  • Monitor and stress-test continuously, adapting governance as model behavior changes rather than relying on point-in-time audits. section 1.3
  • Document model limitations and log inputs and outputs to support transparency. section 1.3

Sensitive information in generated output

Access controls may restrict a user’s permissions while a model still generates output that exposes sensitive information. Governance must therefore consider the output itself, not just whether the user had permission to access a conventional asset. section 1.3

Figure 1.3 Classic GRC compared with AI GRC and GenAI GRC

Compare the classic, AI, and generative AI governance layers that adaptive risk management builds on. section 1.3

How AI Governance covers it

How deep it goes
Contrasts traditional and adaptive governance and identifies monitoring, stress testing, logging, and documentation practices.
Examples
Uses sensitive information leakage despite traditional permissions to illustrate uncertain model-output risks.
What you can do after
Evaluate generative AI governance practices against the need for adaptive controls and continuous oversight.
How it connects
Connects risk management to access control, privacy, prompt injection, and governance, risk, and compliance.

Related concepts

This explainer was generated by Manning's knowledge graph from the text of AI Governance. The section links point to the parts of the book it draws on.