Yesterday I just wrote: After AI agents enter the economic system, the issue isn’t just about how smart they are, but about identity, permissions, and responsibility. Today @Boltzhq provided a counterexample. This non-custodial Bitcoin swap service announced an indefinite suspension. The reason isn’t a one-off incident. Over the past few months, they’ve been repeatedly targeted by automated, AI-assisted infrastructure probing and attacks. All attacks were contained, and no user funds were lost. But the speed at which attackers iterate has already surpassed the ability of a small team to detect and patch vulnerabilities. So the first real change brought by AI×Crypto might not be: Agents helping people make money. But attackers using agents to reduce trial-and-error and attack costs to nearly zero. In the future, an agent capable of managing wallets will at least need: - Spending limits - Address whitelists - Delayed settlements - Emergency shutdowns - Complete audit logs - Clear refund policies and accountability Crypto has given AI the ability to act and make payments. The next step that must be addressed is whether the system has the authority to say “no” to AI in time. An agent wallet without spending limits, shutdown mechanisms, and layers of accountability isn’t part of an economic system. It’s just an automated vulnerability.