Elizabeth Blosh-Myers, VP Third-Party and Model Risk Management at Thread Bank, warns that traditional vendor contracts are struggling to address escalating cyber threats, AI autonomy and supplier concentration. She explains why stronger accountability, continuous monitoring and more adaptable contractual protections will become essential as banks confront increasingly interconnected third-party risks through 2030.
Banking institutions face an increasingly complex third-party
risk landscape as cyber threats, artificial intelligence and regulatory
scrutiny expose weaknesses in traditional vendor contracts.
Yet the greatest challenge may not be identifying emerging
risks, but ensuring that contractual protections are sufficiently practical,
enforceable and adaptable to address them.
Elizabeth Blosh-Myers, VP Third-Party and Model Risk
Management at Thread Bank, argues that effective vendor oversight requires a
fundamental shift away from static contractual arrangements toward mechanisms
that establish clear accountability and support continuous risk monitoring.
Her starting point is straightforward: contractual
protections offer little value if vendors refuse to accept them or institutions
cannot enforce them.
"Vendors will not agree to unlimited liability. They
will agree to clear rules," she says.
Cyber Accountability Requires Practical Protections
For Blosh-Myers, effective cybersecurity provisions begin
with clearly defined incident notification requirements.
She advocates requiring vendors to report suspected incidents
within 48 hours and confirmed incidents within 24 hours, with precise
definitions establishing when those reporting obligations begin.
"Define 'know' or their investigation becomes my
clock," she warns.
Rather than demanding unlimited liability, banks should
negotiate higher liability caps specifically covering security breaches and
regulatory fines.
Subcontractor accountability is equally important. Vendors
should be required to impose equivalent security obligations on their own
suppliers, while banks should retain the ability to reject proposed
subcontracting arrangements.
"Most of the breaches I see come from the vendor's
vendor," Blosh-Myers says.
Contracts should also establish obligations to share
investigation findings, cooperate with regulators and cover notification
expenses when incidents occur.
Her emphasis is on provisions that vendors will sign and
banks can realistically enforce.
"A clause they won't sign protects no one. A clear
trigger and a fix-it period is what I can actually use."
Regulators Want Evidence of Active Oversight
Regulatory expectations are also changing the purpose of
third-party contracts.
Blosh-Myers observes that supervisors increasingly expect
institutions to demonstrate ongoing vendor oversight rather than merely produce
evidence that contractual agreements exist.
"Examiners used to ask if I had a contract. Now they ask
if the contract lets me prove I'm watching the vendor."
This shift places greater importance on detailed audit
rights, including the information banks can obtain, the frequency of access and
arrangements for regulatory examination.
Operational resilience provisions must similarly move beyond
general assurances.
Contracts should establish measurable recovery times, require
evidence of business continuity testing and provide workable exit arrangements,
including the return of bank data in usable formats.
Blosh-Myers points to the Synapse experience as a warning for
banking-as-a-service providers about the consequences of inadequate exit
planning.
Ongoing monitoring obligations should require vendors to
supply risk metrics and complaints information while notifying banks about
ownership changes, offshore relocations and replacements of critical
subcontractors.
She also identifies Tennessee law and courts as a
nonnegotiable contractual requirement in her approach.
Ultimately, contractual language must support demonstrable
controls.
"If the contract can't produce the evidence, the
examiner assumes the control isn't there," she says.
AI Is Exposing Dangerous Contractual Gaps
Artificial intelligence introduces another layer of
complexity, particularly when vendors incorporate AI capabilities into products
governed by agreements originally designed for conventional software.
"Vendors are selling AI and contracting like it's
regular software," Blosh-Myers warns.
She identifies restrictions on vendor use of customer data
for model training as an immediate priority, including prohibitions covering
supposedly anonymized information.
Banks also need sufficient documentation to validate
third-party AI models and contractual notification requirements when those
models change.
Another concern is the introduction of AI functionality into
existing products without customers' knowledge or approval.
Blosh-Myers recommends defining AI explicitly within
agreements and requiring notification before new capabilities interact with
bank data or customers.
Liability presents a particularly significant challenge when
automated systems influence fraud detection, lending or dispute resolution.
Vendors may seek to disclaim responsibility for incorrect AI
outputs, but the bank can still face regulatory consequences.
Contracts therefore need measurable quality commitments,
human review rights and appropriate protections when vendor models cause
compliance violations.
"The bank owns the outcome no matter who owns the model.
Write the contract that way," she says.
Dynamic Contracts Could Transform Vendor Oversight
Looking ahead, Blosh-Myers expects technology to enable
contracts to become active components of third-party monitoring rather than
documents consulted primarily during disputes.
"Yes. The technology is here. The contract language is
what's behind," she says.
She envisages systems linking continuous vendor monitoring
directly to contractual obligations.
For example, deteriorating cybersecurity scores or expired
assurance reports could automatically initiate remediation periods.
Vendor attestations could also be supplied as structured
data, allowing reporting processes to become increasingly automated.