DEFS

DEFS Learn · AI agents and online banking

Can a smaller credit union get managed fraud monitoring for AI agent activity?

Last updated · DEFS Agent Defense (defs.ai)

Short answer

Yes. A managed service can watch the AI-agent channel for you: automated checks on every agent request, human reviewers for the cases that need judgment, real-time alerts for critical events and a monthly outcome report. Your team keeps the rules, the kill switch and the audit trail, without staffing another dashboard. DEFS Agent Defense works this way for credit unions and community banks.

Why smaller institutions struggle with a new fraud channel

At a credit union of a few hundred million to a few billion dollars, fraud, BSA and digital banking often sit with a handful of people. A new channel such as members' AI agents arrives on top of that work. The largest banks are building their own controls; JPMorgan said it is "building tools" while it keeps outside agents out (Banking Dive). Most credit unions need the protection without a build project or a new hire.

What "managed" should mean

  • Automated checks on every request, running around the clock.
  • Named human reviewers for edge cases, working inside authority your institution signs off on.
  • Clear coverage outside review hours: anything unclear is held, never allowed by default.
  • Results you can use: real-time alerts for critical events and a monthly report ready for the board.
  • Control stays with you: your limits, your kill switch, your audit trail.

Questions to ask any managed provider

  1. What can your reviewers decide alone, and what comes to our team? Who signs that off?
  2. What happens to an unclear request at night or on a weekend?
  3. What do we receive each month, and what triggers a real-time alert?
  4. Can we stop all activity ourselves, instantly, without calling you?
  5. Can we export every decision and reviewer action for examiners?
  6. Can we start on test data in a sandbox?

How DEFS handles it

DEFS Agent Defense is a done-for-you service for the AI-agent channel:

  • You set the rules in one kickoff session. DEFS recommends limits; you approve them.
  • Every agent request passes through DEFS before money moves. In-policy actions from verified agents go through, new payees and destinations are held for the member, and forged credentials, unknown payees, over-limit amounts and known rogue agents are stopped.
  • A named DEFS reviewer handles step-ups and edge cases within the authority you delegate, and escalates the rest to you. Outside review hours, anything unclear is safely held.
  • Real-time alerts for suspected rogue-agent campaigns, blocklist hits and new agent types, by email, text or Teams/Slack.
  • A monthly Agent Outcome Report: attempts by agent and credential, what was allowed, held and stopped with dollar amounts, top reasons and time to every human decision.
  • Quarterly policy review with your fraud lead, plus your own kill switch and a tamper-evident audit trail you can export any time.
  • Every engagement starts on test data in a sandbox.

The demo results

We tested DEFS in a model credit union built from public information, using test data. OpenAI and Claude agents sent it 3,259 requests: everyday member tasks mixed with deliberate attacks.

  • 1,490 of 1,491attack attempts stopped by DEFS. The one remaining request moved $0 because the core system rejected it.
  • 0 of 807legitimate requests wrongly blocked: zero false positives.
  • 3,259requests from OpenAI and Claude agents, decided in a median of 7 ms.

"Stopped" means DEFS denied the request outright or held it until the member or a reviewer decided. No attack attempt moved any money.

Attack type in the testResult
Requests sent after the kill switch was turned on71 of 71 stopped
Agents flagged at another institution (shared blocklist)101 of 101 stopped
Requests held for member or reviewer approval, across all test types247

See it for yourself in the live demo environment (test data only), or book a 30-minute walkthrough.

See DEFS stop a rogue AI agent before it moves money, live, in 30 minutes.

Book a demo

Frequently asked questions

Do we have to run a dashboard?

No. DEFS runs it day to day. Your team has full read access, its own kill switch and audit export, and reads the monthly report and answers the rare escalation.

What happens at night and on weekends?

The automated controls run 24/7. Anything that needs a human outside review hours is held, or denied with a member notification, as you choose, and a reviewer picks it up at the start of the next review window.

Does this replace our existing fraud monitoring?

No. It covers the AI-agent channel and works alongside your current tools. Decisions can be exported to your SIEM or case tools, and DEFS escalates into your existing incident process.

How do we get started?

With a kickoff session and your vendor review. DEFS starts in a sandbox on test data, then plans the production connection with your team and your digital banking provider.

Sources

  1. Banking Dive: JPMorgan CEO Dimon warns customers of 'liability transfer' with AI (Oct 13, 2026)
  2. DEFS demo results: model credit union built from public information, using test data (3,259 agent requests from OpenAI and Claude agents).