In partnership with
Good morning, risk and compliance leaders. Happy new month, we won’t talk politics, but dare I say all eyes are on the midterm elections. Last week the FCA set out what good AI governance looks like and firms are scaling carefully, with only 2% running agents at broad autonomy. It was also a week where AI agents were on the news. AI agents are acting for customers faster than anyone has agreed who answers for them. Ten of North America's largest banks have shut the door, and supervisors in Frankfurt and Seoul are treating AI-driven cyber attack as a present problem.
In this week's Accountability Record:
🇺🇸🇨🇦 Ten big banks block personal AI agents, and the liability question lands on customers
🇪🇺 ECB's AI-cyber action plans were due Saturday; the horizontal review starts now
🇰🇷 Korea moves from AI-hack triage to remediation plans and on-site inspections
🇺🇸 CFTC chair asks whether markets need "know your agent"
🇺🇸🇨🇦 US & Canada | Banking, wealth | Security, accountability
Banks block personal AI agents, and the liability question lands on customers
Photo by Atlantic Money on Unsplash
The gap: Major banks are blocking third-party AI agents from customer accounts because nobody can yet verify an agent or say who pays when one goes wrong.
The details:
RBC said on 27 October it will block third-party personal AI agents; existing connections "will stop working". BMO, TD and CIBC confirmed blocks, Scotiabank will restrict them (Canadian Press).
American Banker reports JPMorganChase, Citi, Bank of America, Wells Fargo and Morgan Stanley block external agents.
Chase's terms (effective 27 October) bar "External AI" from using login credentials; customers "assume the entire risk" of improper use.
Wells Fargo terms effective 24 November say AI tool requests may be treated as coming directly from the customer.
Why it matters: Blocking is a stopgap, and a liability clause does not make a complaint go away. Likely owner: the head of digital channels for the access decision, with the CISO for detection and the conduct risk lead for the terms.
Across regimes: US Regulation E treats transfers by someone the consumer gave access to as authorised until the bank is notified; in the UK, such terms would be tested against the Consumer Duty.
Do now: Document your position on customer AI agents, and whether fraud monitoring can tell an agent session from a customer.
The Dell Pro with built-in intelligence.
Built for business. Ready for anything.
With built-in intelligence, you can stay productive and focused on growing your small business. So no matter what the day throws at you, your Dell Pro can handle it.
🇪🇺 EU | Banking | Security
ECB's AI-cyber action plans were due Saturday, and the horizontal review starts now
The gap: Every euro area significant institution owed its supervisor a plan for AI-enabled cyber attack by 31 October, and the ECB will now compare them.
The details:
Letter of 7 July 2026 (SSM-2026-0301) from Claudia Buch, Chair of the Supervisory Board, to the CEO of each significant institution.
Plans go to Joint Supervisory Teams with control measures, resources, roles and timelines.
The ECB "will conduct a horizontal analysis" and share conclusions. The IT Risk Questionnaire moves to February 2027.
Why it matters: Peer comparison turns a plan into a benchmark, so vague ownership will stand out. Likely owner: the CISO for delivery, with the management body member responsible for ICT risk under DORA accountable for the plan.
Across regimes: Builds on DORA's ICT risk framework; compare Singapore's MAS Technology Risk Management Guidelines.
Do now: Check every action in your plan has a named owner and date. Outside SSM scope, run the exercise anyway.
🇰🇷 South Korea | Banking, insurance, securities | Security
Korea moves from AI-hack triage to remediation plans and on-site inspections
The gap: After breaches suspected to involve AI agents, Korean supervisors want written fixes and are sending inspectors.
The details:
The FSC confirmed on 14 October that 16 financial firms were breached since the start of the month, with about 100,000 records exposed.
At an FSS-convened meeting on 30 October, firms were told to submit plans closing gaps found in self-assessments; inspections start with the worst hit.
The FSC meets heads of the five major banks on 5 November on AI-driven hacking.
Why it matters: This is the first supervisory cycle built around a live AI-assisted attack on banks, and the template will travel. Likely owner: the CISO, with the CRO for the risk acceptance decisions.
Across regimes: Mirrors the ECB request above and the Hong Kong SFC's circular on AI-enabled cyberattacks.
Do now: Ask for an inventory of every internet-facing system, including staff tools, with an owner against each.
🇺🇸 US | Derivatives, crypto | Regulation
CFTC chair asks whether markets need "know your agent"
The gap: The derivatives regulator is asking who registers, and who is liable, when an AI agent trades.
The details:
Chairman Michael Selig opened the CFTC's inaugural Frontier Forum, on super intelligence and agentic trading, on 28 October.
He asked: "Does the developer of that agent become subject to registration? Who is liable if the agent goes rogue?"
And: "Do we need a new flavor of know-your-customer, maybe KYA or know-your-agent?" No rules or guidance were proposed.
Why it matters: A chair's questions tend to become examination themes. Likely owner: the chief compliance officer, with the head of electronic trading.
Across regimes: EU and UK MiFID algorithmic trading rules already require algorithm identification and kill functionality.
Do now: List every agent or algorithm that can submit orders, and the person accountable for each.
See what's really happening in every deal
Aligned shows you what buyers are doing between meetings inside your most important deals then surfaces the risks, openings, and next steps. No more guessing where a deal stands or finding out a champion went quiet too late. Just real visibility, so you can act before it's too late.
🛠️ Practitioner's corner: write your customer AI agent position in a week
Find the agents already there. Ask fraud and digital teams for automated or headless sessions in the last 90 days.
Pick a stance per channel. Block, allow via an approved interface, or monitor. Write down why.
Read your own terms. What do they say about credential sharing, and would that survive a complaint?
Name the owners. Access decision, detection, customer outcomes. Record them.
Rehearse one incident. An agent makes a payment the customer disputes. Who decides the refund, on what evidence?
⚡ Quick hits
UK & Europe
Americas
🇺🇸 SEC enforcement director David Woodcock told Reuters the agency is examining AI disclosures: "Our principal concern is that people are making accurate disclosures." Reuters via KFGO
Global
🌐BIS Bulletin 140: AI infrastructure is increasingly debt-funded through off-balance-sheet vehicles, and banks often lend to both vehicles and investors. BIS
🧰 RegTech & security watch
ARTEX, an open-source AI penetration-testing agent, was linked by CrowdStrike to the Korean bank attacks. Control risk: skilled attack tooling now runs unattended, so quarterly patch cycles look slow. BleepingComputer
NatWest reports cutting a compliance review bottleneck by 80% with Adclear's AI. Control risk: faster approvals still need a record of who signed off each output. FinTech Global
📅 Calendar
5 November | Korea FSC meets heads of five major banks on AI-driven hacking
16 November | US interagency third-party risk management guidance: comments due (Fed)
18 November | FCA GC26/7 decentralised finance guidance consultation closes
24 November | Wells Fargo updated online access agreement, including AI tool terms, takes effect
📊 Pulse check
What is your firm's current position on customers' personal AI agents accessing accounts?
Results shared next week.
That’s it for the week!
If something landed, reply and tell me. If something was wrong, reply and tell me that too. I read every response.Forward this to one person who should be reading it.
For sponsorship enquiries, email [email protected]
Until next week.
Emeka Ogbonnaya
Founder BuildMoat