AI Agents That Move Money: How to Build Spending Limits and Circuit Breakers
Table of Contents
Table of Contents
Share

AI agents that move money need limits enforced below the model: one AI bot lost 55.5 ETH, worth $106,200, after a dashboard breach. Build caps and breakers.
Frequently Asked Questions
- No. A rule written in the prompt is an instruction the model can be talked out of, and research on Web3 agents found that prompt-injection defenses offer limited protection once stored memory is corrupted. Enforce the limit in a policy service, signer or contract that the agent calls but cannot modify.
- A spending limit caps how much value an agent can move per transaction and per period, and it lets payments continue inside that cap. A circuit breaker stops all value-moving actions when a trip condition fires, such as repeated policy denials or an unknown counterparty, and stays open until a separate human-held key resets it.
- Fixed-window allowances reset on a schedule rather than looking back over the last full period. An agent that spends the whole cap right before a reset can spend it again right after, so twice the daily figure leaves in minutes. A rolling log or a correctly sized token bucket closes that gap.
- No rule names AI agents directly. EU strong customer authentication rules already pair per-payment amounts with cumulative or count ceilings before re-authentication, and US Regulation E caps consumer liability for unauthorized transfers but excludes transfers by someone the consumer gave access to, which makes credential handover a legal question as well as a technical one.
Don't Miss What's Next
Subscribe to newsletter
AI agent spending limits
circuit breakers
agentic payments
Safe Allowance Module
ICP-3
Get in Touch
Our team will get back to you within 24 hours.

















