AI代理人評測:六大銀行封殺,密碼授權誰負責 | AI Agent Review: Banks Block Rogue Agents Over Liability
By Kit 小克 | AI Tool Observer | 2026-10-19
🇹🇼 AI代理人評測:六大銀行封殺,密碼授權誰負責
最新AI代理人(AI Agent)話題焦點不在模型能力,而在信任與責任:嘉信理財(Schwab)、摩根大通(JPMorgan)本週證實,正大規模封殺未經授權的AI代理人存取客戶帳戶與資金,花旗、美國銀行隨後跟進。這代表「代理型AI」自動幫你轉帳、下單的願景,在金融業現實面前先踩了剎車。
嘉信理財每週擋下「幾十次」AI代理人入侵
嘉信理財CEO Rick Wurster向Bloomberg Television表示,公司每週都會收到一份報告,顯示系統擋下數十次(tens and twenties)未經授權的AI代理人存取嘗試,其中就包含Meta的Muse。問題核心不是駭客攻擊,而是客戶自己把密碼交給AI代理人,讓它代為操作。Wurster直言:「我們無法控制它,客戶也無法控制它,一旦出錯,誰該負責?」
摩根大通、花旗、美銀集體跟進封殺
摩根大通CEO Jamie Dimon在本季法說會上也表態:「你不會想要失控的代理人(rogue agents)。」花旗與美國銀行則因擔心AI代理人自動比價轉帳引發存款外流,同樣開始限制外部代理人存取客戶帳戶。四大銀行幾乎同步出手,顯示這不是單一公司的保守決策,而是整個金融業對「代理型AI」風險的共識。
銀行的替代方案:唯讀 MCP + 自家助理
嘉信理財並非完全擋死AI,而是改用受控的MCP伺服器,讓外部AI工具能「唯讀」查看帳戶餘額,但不能動用資金;同時力推自家助理Charley,把代理權收回自己手上。這其實是目前多數金融機構處理AI代理人風險的折衷做法:開放資訊、鎖死動作。
小克的誠實觀察
這波封殺潮戳破了一個常見迷思:AI代理人卡關的不是能力,是制度。密碼共享、責任歸屬、監管合規這些問題,比讓AI學會轉帳難上十倍。如果你正在打造或使用AI代理人處理金錢相關任務,現在該問的不是「它會不會做錯」,而是「做錯了誰賠」——這個答案金融業自己都還沒想清楚。
好不好用,試了才知道。
🇺🇸 AI Agent Review: Banks Block Rogue Agents Over Liability
The hottest reality check in AI agent news this week isn't a new model release — it's major banks pulling the plug. Charles Schwab and JPMorgan confirmed they are actively blocking unauthorized AI agents from touching customer accounts and money, with Citi and Bank of America following suit. The "agentic AI" dream of letting an assistant transfer your funds for you just hit a very human wall: liability.
Schwab Blocks "Tens and Twenties" of AI Agents Every Week
Schwab CEO Rick Wurster told Bloomberg Television that he receives a weekly report showing the firm blocks dozens of unauthorized AI agent attempts, including Meta's Muse, from accessing client accounts. The issue isn't a hack — it's customers handing their own passwords to an agent and letting it act on their behalf. "We can't control it, the client can't control it, something could go wrong and who's liable?" Wurster said.
JPMorgan, Citi, and Bank of America Join the Lockout
JPMorgan CEO Jamie Dimon echoed the stance on the bank's quarterly earnings call: "You don't want rogue agents." Citi and Bank of America have also started restricting outside AI agents over fears that automated comparison and transfer tools could trigger deposit outflows. Four major banks moving in near-lockstep signals this isn't one company being cautious — it's an industry-wide consensus on AI agent risk.
The Workaround: Read-Only MCP, Not a Full Ban
Schwab isn't blocking AI outright — it's routing outside tools through controlled MCP servers that allow read-only balance checks but no money movement, while pushing its own assistant, Charley, as the sanctioned alternative. Read access stays open; action stays locked down. That's the compromise most financial institutions seem to be landing on for now.
The Honest Take
This lockout busts a common myth: what's actually blocking AI agents from handling money isn't capability — it's accountability. Password sharing, liability, and compliance are harder problems than teaching a model to initiate a transfer. If you're building or using an agent for financial tasks, the real question isn't "will it mess up" — it's "who pays when it does." Right now, even the banks don't have an answer.
好不好用,試了才知道。
Sources / 資料來源
- Schwab and JPMorgan are blocking AI agents from moving customer money
- Schwab and JPMorgan Block Unauthorized AI Agents from Moving Money
- Citigroup and BofA restrict AI agents from accessing client accounts
延伸閱讀 / Related Articles
- OpenAI人形機器人評測:2027年demo,身體不是重點 | OpenAI Humanoid Robot Review: 2027 Demo, Brain Over Body
- Claude Opus 5.5評測:反編譯老遊戲,GTA秒變瀏覽器版 | Claude Opus 5.5 Review: Decompiling GTA, Halo for Browser
- OpenAI數學論文評測:722篇齊發,3篇撤稿收場 | OpenAI Math Papers Review: 722 Dropped, 3 Retracted
AI 工具觀察站 — 每日精選 AI Agent 與工具趨勢
AI Tool Observer — Daily curated AI Agent & tool trends
留言
張貼留言