Search papers, labs, and topics across Lattice.
This paper quantifies the economic risks associated with autonomous AI agents by introducing a novel method called trace-economic underwriting, which maps tool-use traces to customer exposure and claimable losses. The approach allows for the pricing of risks in a way that makes automation economically viable, as it ensures that the expected benefits of deploying AI exceed the associated costs and risks. Key findings include a significant reduction in pricing mean absolute error from $17.7K to $569 and a 72% reduction in conditional value-at-risk, demonstrating the method's effectiveness in managing and transferring risk.
Trace-economic underwriting could revolutionize how we assess and insure against the risks of autonomous AI, making automation financially viable.
AI agents can now take irreversible actions in operational systems, but agent-caused losses are still not clearly assigned, priced, or transferred. Providers often disclaim consequential damages, users are left with uncompensated losses, and default human review limits the efficiency gains of automation. We ask when autonomous AI deployment can become economically acceptable despite failure risk. Our answer is to quantify risk at the customer-task-trace episode level and transfer it through insurance. Automation is acceptable when its expected benefit exceeds the premium, control cost, and remaining risk. This requires a defined role with bounded permissions and comparable traces. We introduce trace-economic underwriting, which maps tool-use traces to customer exposure and claimable loss, then uses this representation for pricing, control, and risk transfer. It uses deterministic economic labels rather than an LLM judge. In our trace-to-loss testbed, trace-economic pricing reduces pricing MAE from $17.7K to $569 and removes regressive cross-subsidy. A 300-trace expert audit accepts 295 labels unchanged. On 1,000 real SWE-smith traces, trace-conditioned controls reduce CVaR95 by 72%. Theorem~1 gives a finite-sample scope condition. We release code, labels, and audit sheets.
