AI Agents Now Have a 'Spending Limit' to Protect Your Wallet

Fleets of LLM agents can cause irreversible harm—moving money, deleting data, or disclosing information—and current per-effect checks let individually authorized agents collectively overdraw their principal's risk while every local gate stays correct. The proposed “irreversibility budget” is a cumulative account of residual value-at-risk, maintained by a trusted runtime for each principal across agents, workflows, and tenants; the runtime charges each effect and denies the marginal one when the aggregate would overdraw. In a controlled study, per-effect gates admitted fleet-level overdraws up to 48 times the tenant's risk limit, while the budget held every correctly charged run within that limit—though getting the price right, with heterogeneous and adversarially declared effects, remains the key open problem.