# Agent Payments Widen the Wallet Attack Surface

As AI agents begin to hold and spend stablecoins, the gap between what users approve and what signed transactions actually do becomes a recurring loss channel, forcing wallets and payment standards to add intent verification.

- Conviction: 41 / 100 (strengthening)
- Horizon: Emerging (watchlist)
- Tracking since: 2026-11-20T00:00:00.000Z
- Last updated: 2026-11-22T06:32:00.503Z
- Canonical: https://polylog.news/crypto/trends/agent-payments-widen-wallet-attack-surface
- Publisher: Polylog

## Recent score history

- 2026-11-21: 38
- 2026-11-22: 41

## Recent evidence

- [confirms] Singapore plans a sandbox for AI agents that pay in digital money, and Hong Kong proposes jail for false crypto license claims (2026-11-22): Singapore plans a sandbox for AI agents that pay in digital money, testing how banks confirm an agent's authority before it pays. A regulator treating agent authorization as the main control is the intent-verification gap the thesis identifies.
- [confirms] Researchers Show AI Agents' Wallets Can Sign Valid Transactions Their Owners Never Authorized (2026-11-20): A new research paper shows that the spending limits users approve for AI-agent wallets do not constrain what the signed transaction actually does on-chain, so agents can produce valid signatures for transfers their owners never authorized. This documents the gap between approval and execution that the thesis expects to become a loss channel.
