Fiduciary resilience · Publication 001

When Financial Records Disagree

The next test of fiduciary resilience

Fraud controls verify people and instructions. Fiduciary resilience must also establish which balance, ownership, valuation, and settlement records can be trusted when a custodian or shared provider is impaired.

Editorial scope. This article proposes a resilience framework. It does not assert that every element is currently required by fiduciary law, regulation, or contract. Duties vary by role, mandate, governing instrument, jurisdiction, and applicable law.

Most fiduciary controls ask whether the right person authorized a transaction. The harder question begins after authorization: if a bank, custodian, fund administrator, or shared technology provider is impaired, can the fiduciary establish what the client owns—and demonstrate which record should be relied upon?

A September 2026 shared-core banking outage offers a useful warning without requiring a cyberattack. Affected credit unions reported unavailable online banking, last-known balance information for some card and ATM decisions, held electronic transactions, limited offline service, and sequential restoration across more than 300 institutions.12 The institutions said the disruption followed equipment and cooling failure—not malicious access. The episode therefore does not prove that financial records were corrupted. It demonstrates how failure at a common provider can leave downstream institutions and clients operating with delayed, incomplete, or outdated information.

AI raises a separate but related concern. A September 2026 Financial Stability Institute paper reports that frontier models can accelerate vulnerability discovery, exploit development, and complex cyber operations, compressing the time available for institutions and providers to respond.3 The underlying resilience problem is longstanding; AI may increase its speed, scale, and severity.

Financial Integrity Watch proposes a practical framework: protect both the authority to act and the truth of record.

60-second summary

Observed

A shared-provider equipment failure disrupted access, delayed transactions, and forced institutions to rely on stale or limited information.

Changing

Frontier AI can compress cyber response windows and amplify existing concentration and recovery risks.

Proposed

Fiduciary diligence should test how reliable financial state will be established when normally trusted records disagree.

Not established

No cited evidence shows an autonomous AI corrupting a major bank ledger, depository, or settlement system.

01 · Proposed framework

Two protections that belong in one framework

Authority to act

Was the person authentic and the action authorized?

  • Client and representative authentication
  • Dual authorization and separation of duties
  • Callback and out-of-band verification
  • Transaction-pattern review
  • Staff training and insider-threat monitoring

These controls remain essential.

Truth of record

Which financial state can be relied upon?

  • Which balance is correct?
  • Who legally and beneficially owns the asset?
  • Which instructions were accepted?
  • Which transactions became final?
  • Which recovery copy can be trusted?

This complements—not replaces—authorization controls.

Authoritative record means the record accepted as controlling for a particular purpose when systems disagree, subject to governing law, contracts, account agreements, and market-infrastructure rules. Clean recovery means restoring operations only after material records and transactions have been checked for integrity, completeness, order, and duplication.

02 · Point / counterpoint

Is this really a new fiduciary obligation?

Point

Banks, custodians, trust companies, advisers, and other financial institutions already maintain controls for custody, reconciliation, books and records, fraud prevention, business continuity, audit, backup, recovery, and third-party oversight. A service interruption does not automatically invalidate legal ownership or mean that records have been corrupted.

Counterpoint

Those controls may not have been tested against a condition in which several apparently independent safeguards depend on the same cloud provider, processor, identity system, software platform, data source, administrator, AI provider, or recovery environment.

The proposed advance is not indiscriminate duplication of every record. It is governance over how reliable financial state will be established when normally trusted records disagree or cannot be promptly verified. Banking guidance already emphasizes integrity across backups and replicas, including the risk that corruption may propagate into recovery environments.4

03 · Why now

What AI changes

AI did not create third-party concentration, reconciliation risk, or the need for trustworthy records. It may change the operating conditions by increasing:

  • The speed of vulnerability discovery and exploitation
  • The scale and adaptability of automated operations
  • The number of machine identities and delegated actions
  • The difficulty of monitoring complex agent activity
  • Dependence on common AI, cloud, and software providers
  • Pressure on detection, containment, and recovery time

Financial-sector research and industry guidance now describe compressed remediation windows and the need to reconsider traditional technology-risk cycles.35

AI is a threat and tempo multiplier—not the origin of financial record-integrity risk.

04 · Illustrative scenario

An upstream integrity cascade

This is a resilience scenario, not a claim that this exact AI-driven event has occurred.

  1. A bank, custodian, processor, or shared provider suffers a serious compromise or disruption.

  2. Internal ledgers, settlement messages, and external transaction records no longer reconcile.

  3. Dashboards and reports may continue to appear coherent because they are generated within the affected environment.

  4. The provider pauses some activity while reliable state is reconstructed.

  5. Downstream fiduciaries experience delayed withdrawals, purchases, redemptions, distributions, or client reporting.

  6. Recovery requires more than restarting applications. Material in-scope transactions must be checked for completeness, duplication, omission, sequence, authorization, and finality.

The crucial question is not merely whether data still exists. It is whether the relevant parties can demonstrate which information should be trusted. CPMI-IOSCO’s guidance for financial market infrastructures similarly emphasizes response, recovery, testing, and coordination; its direct scope is FMIs, but its resilience principles are informative by analogy.6

05 · Board agenda

Four actions for boards and fiduciaries

01

Map authority

For each material asset and transaction type, identify which institution and record carry legal, contractual, or operational authority—and what happens when records disagree.

02

Preserve evidence

Determine which instructions, acknowledgements, positions, journals, and settlement records require sufficiently independent protection, based on law, privacy, architecture, and risk.

03

Define escalation

Establish how conflicting information will be investigated, who must be notified, which activity pauses, and which legal or market authority decides.

04

Prove clean recovery

Test completeness, sequence, effective dates, authorization, deduplication, finality, exception resolution, and external reconciliation before normal processing resumes.

06 · Questions to ask now

Six questions for management and providers

  1. Which institution and record establish the controlling balance, ownership, valuation, and settlement position for each material asset class?
  2. What is the formal process when internal records disagree with those of a bank, custodian, transfer agent, depository, processor, or fund administrator?
  3. Which sufficiently independent evidence would help reconstruct client positions without relying solely on the affected production environment?
  4. Are critical backups, journals, and audit records protected from the same identities, networks, administrators, and providers as production?
  5. Has the organization tested plausible but incorrect information, conflicting records, or compromised recovery data—not merely an unavailable system?
  6. Who is authorized to declare that restored financial records can again be relied upon, and what evidence must support that declaration?

07 · Evidence boundaries

What the evidence establishes—and what it does not

Establishes

  • Frontier AI is increasing cyber capability and compressing remediation windows.
  • Financial institutions depend on concentrated technology and processing providers.
  • Provider outages can produce stale information, held transactions, service restrictions, and sequential restoration.
  • Supervisory and industry guidance emphasizes operational resilience, integrity-aware backups, testing, recovery, and third-party oversight.

Does not establish

  • That autonomous AI has corrupted a major bank ledger, securities depository, or settlement system.
  • That the cited credit-union outage was a cyberattack or involved corrupted records.
  • That this proposed framework is already a universal legal fiduciary duty.
  • That financial collapse is imminent.

Current assessment: Material and rising—not catastrophic. Preparation is warranted; alarmism is not.

Conclusion

From reassurance to demonstrable resilience

A fiduciary organization may authenticate every client and properly approve every instruction while remaining dependent on an upstream institution to establish what the client ultimately owns. That does not make existing controls obsolete. It reveals a complementary responsibility.

  1. Confirm that the person and instruction are authentic.
  2. Demonstrate that the resulting balance, ownership, valuation, and settlement records can still be relied upon.

Whether this becomes “the next fiduciary standard” is ultimately a question for fiduciaries, clients, counsel, regulators, and courts. It should, however, become a practical diligence question now.

A useful first exercise

Select one material asset class. Trace it from client authorization through custody and settlement. Identify the controlling record at every step—and determine what evidence would govern if two normally trusted records disagreed.

Sources and evidence notes

Primary and first-party references

Show the six annotated references
  1. Wayne County Community Federal Credit Union, “Banking System Outage,” updated 17 September 2026. Affected-party notice
    wccfcu.com/banking-system-outage

    Supports the provider outage, offline/last-known balance operation, held electronic transactions, and sequential recovery. Does not establish a cyberattack or record corruption.

  2. Badger Globe Credit Union, “Important Update About Our Banking Systems,” updated 18 September 2026. Affected-party notice
    badgerglobecu.org/important-update-about-our-banking-systems

    Supports more than 300 affected credit unions, equipment/cooling failure, limited current-balance access, transaction delays, and the statement that the incident was not a cyberattack.

  3. Financial Stability Institute, “When machines attack: frontier AI cyber threats and policy responses in the financial sector,” 9 September 2026. Institutional analysis
    bis.org/publications/fsi-paper-28

    Supports frontier-model cyber capability, compressed remediation windows, and amplified third-party dependencies. This is FSI analysis, not a binding rule.

  4. Federal Financial Institutions Examination Council, “Data Backup and Replication.” Supervisory guidance
    ithandbook.ffiec.gov — Data Backup and Replication

    Supports data-integrity controls across production, backups, and replicas, as well as recovery preparation for destructive events. Its direct scope is supervised financial institutions.

  5. FS-ISAC, “Updated Sector Risk Advisory: Preparing the Enterprise for AI-Enabled Vulnerability Discovery,” 22 July 2026. Industry guidance
    fsisac.com — Updated Sector Risk Advisory

    Supports concern that AI-assisted vulnerability discovery is narrowing response windows and requires faster reprioritization. It is sector guidance, not regulation.

  6. CPMI-IOSCO, “Guidance on cyber resilience for financial market infrastructures,” June 2016. International standards guidance
    bis.org/cpmi/publ/d146.pdf

    Supports response, recovery, testing, coordination, and learning for financial market infrastructures. Its direct scope is FMIs; application to other fiduciary organizations is by analogy.

Methodology: Claims are separated into observed events, authoritative or industry guidance, Financial Integrity Watch analysis, and illustrative scenarios. Accessed 19 September 2026. See the site methodology and corrections policy.