The rise of foundation models offers enterprises exciting new possibilities but also raises new and expanded questions about their ethical design, development, deployment and use. According to a recent IBM Institute for Business Value generative AI survey, organizations are already expressing concerns about trust-related issues—specifically as barriers to investment. Their top concerns are cybersecurity (57%), privacy (51%) and accuracy (47%). Many organizations were taking these concerns seriously before the consumerization of generative AI, expressing their intent to invest at least 40% more in AI ethics over the next three years. Awareness about risks and possible ways to mitigate them is the first crucial step toward building trustworthy AI systems.

​

In this document we:

• Explore the benefits of foundation models, including their capability to perform challenging tasks, potential to speed up the adoption of AI, ability to increase productivity and the cost benefits they provide.
• Discuss the three categories of risk, including risks known from earlier forms of AI, known risks amplified by foundation models and emerging risks intrinsic to the generative capabilities of foundation models.
• Cover the principles, pillars and governance that form the foundation of IBM’s AI ethics initiatives and suggest guardrails for risk mitigation.

• Benefits of foundation models
• Risks of foundation models
• Risk Examples
• Principles, pillars and governance
• Guardrails and mitigations
• AI policies, regulation and best practice