The rising enthusiasm around generative AI and the prospective increase in AI adoption are opening up intriguing opportunities for insurers to underwrite AI risks.
Executive SummaryGenerative AI adoption brings new risks, including physical, financial and psychological harm, demanding innovative insurance solutions such as algorithmic liability coverage, write PwC's Anand Rao and Marie Carr. Here, they compare AI risk insurance with cyber insurance, noting that in spite of similarities, the rapid generative AI adoption and broader risk spectrum require more tailored and agile coverage solutions, including AI intellectual property coverage, autonomous vehicle insurance, AI ethics and compliance coverage, in addition to algorithmic liability coverage.
However, before they delve into the AI risk opportunities, insurers need to address three critical questions: What aspects of the AI risks are already covered by existing insurance products (e.g., product liability, cyber risk)? What lessons can be drawn from the evolution of cyber risk insurance over the past few years that are relevant to AI risk insurance? And finally, what is the market for AI risk insurance and how can this risk be assessed and underwritten?
Coverage of AI Risks
AI is seldom used in isolation; it is integrated into existing automation solutions or various business, engineering or scientific applications, or hardware and software products. For legal purposes, AI is treated as advanced or complex software. As such, current cyber risk insurance covers AI-related risks surrounding data breaches, business interruption due to a cyber attack, privacy liability, deep fake cyber extortion, cyber stealing attacks, data leaks, network security liability and notification costs. When AI is incorporated into physical products like autonomous vehicles, robots and industrial machinery, product liability insurance may also apply. This insurance typically covers physical injury and third-party damage caused by design defects, manufacturing process or production flows, and marketing/sales errors.