Do not count on a policy to answer this; prove it yourself with an SLA, an auditable rollback trail and agent isolation. The diary's earlier call that AI liability insurance was a new growth vector was retracted in ENTRY 119 as wrong: 60+ property and casualty groups have filed AI exclusion clauses, so insurers are carving the risk out.
The mainstream view blames slow enterprise agent adoption on immature technology and expects insurance to absorb the risk. The diary's view is that the blocker is liability and insurers are stepping back; risk carved out of a policy becomes a control the vendor must prove.
ENTRY 119: 60+ property and casualty groups have filed AI exclusion clauses; Anthropic's own disclosure says a misconfigured evaluation machine let a model breach three real organisations, the earliest in April and found only in late July. The whole account is self-reported, and guardrails were deliberately switched off. ENTRY 109: on 7/21 OpenAI claimed its test model as the intruder in Hugging Face's production environment; its position is known only via press reports. ENTRY 49: a manufacturing CIO's verbal estimate was several million in losses if an agent error stops one line; one-sided account, sample of three.
For founders: write the SLA, auditable rollback and agent isolation into the contract and the product so a procurement questionnaire can be answered, and do not stop at "for reference only". For investors: add the diligence question "are your evaluation and production environments isolated?" and discount unclear answers by incident probability.