From Exclusion to Affirmative Coverage: Risk Taxonomy of Agentic AI and Insurance as a Private Regulator (Revista de Educação e Pesquisa em Contabilidade (REPeC))

Summary

• The paper concludes that AI is an insurable risk, although systemic tail risks and catastrophic failures of foundation models require compensation funds and/or state-backed alternative systems.p.2
• Affirmative policy language is needed because the ambiguity of silence (silent AI) is a market failure that supervision should correct.p.2
• Mandatory incident reporting is presented as the only route to sufficient and reliable loss triangles.p.2
• The market is moving toward a plurality of models: modular covers for startups, performance warranties, validation-backed insurance, excess/wrap covers and products for agentic AI.p.17

Key insights

1. AI should be treated not as a homogeneous risk block but as a heterogeneous set of exposures whose insurability depends on estimability, mutualization, traceability and technical governance.p.4
2. Agentic AI can combine access to confidential data, exposure to untrusted content and external communication, a lethal triad vulnerable to prompt injection.p.10
3. Intentional acts such as AI-washing fall outside coverage under the classic doctrine that malice is not insurable.p.11
4. Algorithmic bias is insurable only in its fortuitous and controlled form, while intentional discrimination is excluded as contrary to public policy.p.14
5. Concentration of generative AI models in a small list of providers means one failure can hit many users at once, defeating loss independence.p.13
6. Static underwriting must give way to dynamic, predictive underwriting with retro-rating and continuous monitoring for a mutable risk.p.12
7. Munich Re aiSure structures cover around technically defined AI errors and contractual warranties, with a quasi-parametric trigger.p.18