We feel like dealing with AI today, unleashed in the world with excitement, rosy predictions and expectations, is similar to how 20th century physicists handling research around radioactive materials like radium: no actual protection, minimal security and free experimenting. Later, they realized that failure modes are not just bad - they could be deadly. As a result, today almost 10-20% of all the budget allocated for nuclear energy project is dedicated for security, protection and building full proof option to deal with failure modes. We can learn the lessons from nuclear energy industry and preventively start assessing failure modes in AI Identity Management.
In this presentation, we will discuss how AI failure modes (AIFM) can be identified, measured and mitigated during all stages of both “IAM for AI” and “AI for IAM” project types: from initial design to full implementation and day-to-day operations. We would provide argumentation that 10-20% of AI project budget should be allocated towards building guardrails, like access control, identity governance and cybersecurity measures around AI activities, especially Agentic AI, which is very similar to moving from natural radioactivity to artificial one.
