- In 2025, major insurance providers filed regulatory complaints regarding the inclusion of AI-pov’язаних risk factors in corporate policies, and the Insurance Services Office submitted a standard generative AI exclusion.
- The materials note that insurers may be required to cover risks and claims arising from AI, and that this is why insurers are seeking to exclude AI from coverage.
- Some materials on general AI liability include a reference to governance for real-world operational systems.
In 2025, according to TechCrunch, major insurers have moved to exclude AI from coverage, citing risks related to corporate liability. The companies, which include Great American, Chubb, and W.R. Berkley, claim that the share of coverage for directors and officers and errors and omissions may be possible to reduce.
Also, materials indicate that Insurance Services Office’s generative AI exclusion, which insurers may add to coverage for commercial general liability claims.
The materials also note that insurers are likely to face additional risk: insurers cannot eliminate the risk of damage and model failures, but they can reduce the risk by excluding AI from coverage. The author notes that in systems, AI is not only a factor that changes processes, but also a factor that triggers technical and legal consequences, and that the risks may be possible to reduce.
In addition, the materials note that the problem with agentic AI is that it is difficult to predict and control the agent’s actions, and that this is why the agent’s “accountable AI” — as opposed to the “black box” — is required. The materials also suggest that it is important to ensure that the system can be traced and that the system’s actions can be explained, including through document-level traceability, such as CRM or ERP data, rather than through “guesswork,” which might be based on real-world reasoning.
One material also proposes work on agentic AI with a focus on semantic understanding of data and the ability to trace the system’s actions. However, governance can be difficult to implement in practice.