Short · Executive leadership · 22 min read

A Supplier Was Hit by Ransomware: What Should You Do?

If a supplier reports ransomware, establish how its incident could reach your organization, restrict the affected paths, activate continuity arrangements and investigate your own environment in parallel. Restore each connection only when evidence supports its particular use. A supplier's return to service is relevant information; it doesn't establish that your data, delegated access, software deliveries and business transactions are all safe.

I would make the response about dependencies rather than the supplier's overall status. A payroll export, a remote administrator account and an automatic software update create different exposures. They may share a supplier name while requiring different containment actions, different evidence and different people to authorize restoration. One green status against the supplier conceals those differences.

This paper proposes a customer-side decision framework for incomplete information. It synthesizes incident response, supply-chain security, identity and recovery guidance. It is not an empirical study or a guarantee that a particular reconnection is safe. The worked example is fictional. The central question is how to make defensible operating decisions while the supplier is still investigating, without treating either uncertainty or commercial urgency as permission to skip the evidence.

Separate the supplier's incident from your exposure

Begin with what the notification actually establishes. Which legal entity, service, environment and period does it concern? Does the supplier report encryption, unauthorized access, suspected data removal, loss of administrative control or simply a service interruption? Preserve its original wording and distinguish confirmed observations from hypotheses. An early statement can be useful without being complete.

The joint #StopRansomware guide distinguishes encryption from data extortion, including attacks that steal information without encrypting systems (Cybersecurity and Infrastructure Security Agency [CISA] et al., 2023, introduction). That distinction prevents two errors: assuming an available service means confidentiality was preserved, and assuming a ransomware announcement proves that all customer data was stolen. Neither conclusion follows from the label alone.

Build an initial exposure statement with explicit unknowns. For example: the supplier processes a defined dataset; it has a named support identity; no customer-side unauthorized action has yet been established; logs cover a stated interval; the supplier has not confirmed whether its identity systems were affected. That statement is more actionable than saying the organization is unaffected because its endpoints still work.

Identify the evidence that could change each conclusion. If the supplier later discloses administrative compromise, the support identity needs renewed scrutiny even when the original report mentioned only encrypted servers. If customer logs show activity outside the expected scope, investigate it through your incident process. The supplier's investigation and yours should inform each other; neither should become a reason to postpone the other indefinitely.

Map four ways the incident can cross the boundary

I would separate the relationship into four channels. This is an analytical classification, not an official severity scale. Some services occupy several channels simultaneously, and a subcontractor can introduce another dependency behind the visible supplier.

ChannelCustomer-side questionInitial decision to resolve
Data custody and exchangeWhat information is held, received or transmitted?Pause which transfers, preserve which records and assess which exposure?
Delegated authorityWhat can supplier people, tools or identities do here?Restrict which grants, sessions, agents or management paths?
Software deliveryWhat code or configuration can the supplier introduce?Hold which releases and establish which trusted baseline?
Operating capacityWhich business process stops or degrades without the service?Use which fallback, for how long and with what reconciliation?

The channels are related but do not substitute for each other. Disabling a support account doesn't retrieve records already held by the supplier. Pausing an update feed doesn't stop an established remote-management session. A manual workaround can preserve an essential service while accumulating transactions that later need careful reconciliation. The response should make those remaining conditions visible.

For managed service providers, the joint advisory specifically recommends reviewing customer-provider connections, managing identity-provider trusts, separating networks and avoiding administrative credential reuse across customers (National Cyber Security Centre [NCSC] et al., 2022, pp. 4–5). Those recommendations give concrete starting points for the authority channel. They don't establish that every supplier incident has crossed every connection.

Assign an owner to each consequential channel and identify its actual enforcement point. Procurement may own the contract while a platform team controls an application grant and an operational team owns the service interruption. A relationship manager cannot promise containment that another team has not implemented. Conversely, an administrator should not silently accept a business outage beyond their delegated authority.

Coordinate containment and continuity immediately

“Disconnect everything” is too coarse to be the default instruction for every supplier. The relevant decision is what must be stopped now, what can be constrained and what operational harm the restriction creates. A credible path for ongoing compromise calls for urgent containment. A supplier supporting a safety-critical process also requires the relevant operational authority to assess how interruption can be performed safely.

The #StopRansomware response checklist recommends coordinated isolation of affected systems and warns that powering down systems can destroy volatile evidence (CISA et al., 2023, response checklist, steps 1–2). That is guidance for responding to affected environments, not a universal rule to turn off every customer connection after any vendor announcement. Apply its containment and evidence-preservation principles to the paths actually under investigation.

Work in parallel: technical responders restrict exposure and preserve records; service owners activate tolerable alternatives; the supplier contact seeks decision-relevant information; the designated incident authority resolves conflicts. Record the affected service, action, effective time, expected impact and way to reverse or replace the action. Use an independently verified communication channel if the supplier's usual mailbox or support portal may be affected.

A temporary restriction needs a meaningful scope. Blocking interactive support login may leave machine-to-machine access available. Allowing read access may still permit confidential extraction. “Read-only” therefore describes an authorization property, not a complete confidentiality argument. State which consequence the restriction reduces and which consequence remains possible. Where a necessary path cannot be restricted, name the unresolved exposure and escalate the operating decision rather than recording containment as complete.

Prove what happened to delegated access

Inventory the capabilities the supplier could exercise: named accounts, service principals, API credentials, federation trusts, remote-management tooling, local credentials and recovery mechanisms. Use actual grants and configuration records, not only the contract's description of support. Identify the resources and actions those capabilities reach, including access that survives a normal user sign-out.

NIST's federation guidance explains that relying parties manage sessions separately from identity-provider sessions. Terminating an identity-provider session does not necessarily terminate downstream sessions, and assertion validity is distinct from session lifetime (Temoshok et al., 2025, §4.7). Consequently, disabling a supplier's central login should not automatically close the question of existing application access. Determine which downstream mechanisms require separate invalidation or restriction.

An administrative success response is also limited evidence. Under OAuth token revocation, HTTP 200 can mean successful revocation or that the submitted token was already invalid (Lodderstedt et al., 2013, §2.2). The response does not enumerate every credential the supplier holds or demonstrate denial at every protected resource. It supports a narrower conclusion about that request.

I would pair the authorized change with a bounded verification at the relevant resource, using approved test identities and methods. Record the identity, action, resource, enforcement layer, observation and time. If a historical session cannot be reproduced, state that limit and identify the server-side evidence or compensating restriction available. Don't retain a suspected exposed credential in a casual test document. The objective is evidence of the boundary, with credential handling controlled by the incident team.

Establish a separate software acceptance decision

A supplier can distribute code as well as operate a service. CISA and NIST describe software supply-chain compromise as interference before software reaches the customer, including compromise introduced through patches or hotfixes (CISA & National Institute of Standards and Technology [NIST], 2021, introduction). A supplier's restored ticket portal therefore says little about a release produced during the incident window.

Identify installed versions, pending packages, automatic update channels and administrative configuration feeds associated with the affected service. Preserve package identifiers and relevant installation records. A pause in new deployment may be justified while the trustworthiness of the delivery process is assessed, but it should receive an owner and review point: indefinitely freezing updates can also leave unresolved exposure. The choice depends on the actual release, threat and available alternatives.

NIST's Secure Software Development Framework recommends providing mechanisms to verify release integrity and protecting release artifacts and provenance information (Souppaya et al., 2022, PS.2–PS.3). Integrity verification is necessary evidence, but its meaning depends on the reference and authority being trustworthy. Matching a package to a hash supplied through the same potentially compromised channel establishes agreement with that reference; it doesn't independently establish a clean build.

Ask what supports the particular version proposed for use: the build environment's assessed scope, protected release records, handling of affected signing authority and the supplier's account of changes since the incident. These are questions, not allegations that a signing key was compromised. A verified replacement support account should not authorize deployment of an unassessed software package. Keep software acceptance distinct until its own evidence is adequate.

Ask for evidence that can change a customer decision

A demand for the supplier's entire forensic report may be impractical and unnecessary. Start with the decision: whether to transmit a dataset, allow a support operation, install a release or resume transaction processing. Then specify the minimum information needed to resolve that decision and what remains uncertain if it cannot be supplied.

Useful requests identify service and tenant scope, relevant dates, the mechanisms affected, customer actions recommended, credential or release implications, retained evidence and the next update condition. Ask the supplier to distinguish “not observed” from “excluded by the investigation.” A negative conclusion based on incomplete retention or inaccessible systems has a different evidential strength from one supported by coverage of the relevant path and period.

NIST's supply-chain guidance addresses documenting incidents and response decisions and establishing protected, approved communication and escalation arrangements with suppliers (Boyens et al., 2022, IR-5–IR-6). Use those arrangements to share necessary information with authorized recipients. Don't send sensitive incident material to an unverified contact merely because the request appears urgent or includes a familiar supplier logo.

The resulting evidence may be a scoped attestation, a technical briefing, selected records or findings reviewed under appropriate confidentiality arrangements. Judge the content, provenance, scope and unanswered questions rather than the document's title alone. An independent investigator's involvement can add useful scrutiny, but the presence of that investigator does not mean every customer channel has been examined or approved for reconnection.

Investigate from your side when supplier visibility is limited

Preserve relevant customer-side evidence before ordinary retention removes it: access and administration records, application grants, transfer histories, deployment records and business transaction states. Establish a timeline with source timestamps and known clock or collection gaps. A customer-visible event may be the first indication that the supplier's initial scope statement needs revision.

Forensic guidance emphasizes identifying potential data sources and recognizing that useful evidence may be controlled by another organization; alternative sources may be needed when a primary source is unavailable (Kent et al., 2006, §3.1.1). I would apply that principle by asking what the customer's systems can establish independently. A local audit trail might confirm a privileged change even when the supplier cannot yet provide its operator logs.

Independent observation still has limits. The absence of a logged export does not establish that the supplier-held copy was untouched. A clean endpoint scan does not settle whether an API credential was used elsewhere. Document coverage by resource, action and period. Where an investigation depends on records that no longer exist, retain the uncertainty and assess the operating implications instead of translating missing evidence into reassurance.

Keep technical findings separate from notification decisions. NIST's incident-response guidance calls for established coordination procedures and notifications appropriate to applicable legal, regulatory and contractual requirements (Nelson et al., 2025, RS.CO-02). It supplies no universal deadline for this scenario. Have the designated legal and contractual owners assess the actual facts and obligations. Supplier notification to you does not, by itself, settle what you must communicate to others.

Treat business continuity as a changing constraint

A supplier recovery estimate is not the customer's demonstrated ability to continue operating. Map the process affected, current work already accepted, upcoming commitments, acceptable degradation and dependencies of the fallback. A substitute service that relies on the same identity provider, data feed or subcontractor may not remove the relevant failure condition.

NIST's contingency-planning guide distinguishes maximum tolerable downtime for a business process from recovery time objectives for supporting system resources (Swanson et al., 2010, §3.2.1). The source is federal guidance; I use the distinction analytically rather than importing its historical legal requirements. A technical restoration target must fit inside the organization's actual operating tolerance, including the work needed after systems become available.

For example, a supplier could restore an application before the organization has validated data, reconciled manual transactions or enabled its staff to resume work. Those activities consume time and capacity. Record them explicitly when assessing the fallback. A manual process may work at a modest transaction rate but become unsafe as queues grow, staff fatigue increases or authorization checks are bypassed. Those are conditions to investigate, not assumed universal effects.

Review continuity against observable triggers: an approaching commitment, an unreconciled queue, unavailable specialist staff or failure of a compensating control. Assign someone to report when the approved operating envelope no longer holds. “Continue manually” should include who can authorize the work, where the authoritative record is kept and what will trigger a different decision. Otherwise the workaround becomes another unexamined dependency.

Reconnect by capability, with explicit acceptance and withdrawal

I would use four dispositions for each channel: blocked, limited operation, approved for the stated scope, or unknown. Unknown is an evidence state, not a reason to mark the channel safe. Limited operation must name what is permitted, who accepts the remaining exposure, how activity is observed and what condition ends the permission. This is a proposed record format, not a validated risk score.

NIST's recovery guidance recognizes that recovery may be partial and that restoration decisions balance operating needs with investigation and the risk of leaving attacker access behind. It also discusses establishing recovery criteria and authority before the event (Bartock et al., 2016, §§2.3.3–2.3.4). That supports a controlled resumption decision under stated uncertainty; it doesn't justify reconnecting merely because a supplier requests it.

For each proposed capability, assess four distinct questions: is the supplier-side condition sufficiently understood for this use; has the customer's boundary been configured as intended; can the business verify the resulting operation; and can permission be withdrawn if new evidence changes the assessment? A failure in one necessary condition cannot be compensated for by a reassuring answer to an unrelated one.

Proposed useEvidence to examineWhat the decision does not authorize
Receive a bounded data exportAuthenticated source, expected scope, integrity and reconciliation checksStanding remote administration or unrestricted uploads
Permit a support taskApproved identity, limited resources/actions, observation and removal testPermanent privileged access or other support identities
Deploy a replacement releaseAssessed release provenance, integrity and controlled acceptanceOther versions or automatic future releases
Resume transaction exchangeAgreed boundary state, tested processing and duplicate/missing-item handlingBlind replay of every queued operation

These examples require adaptation to the actual platform. Some services cannot be separated into the proposed capabilities. If technical granularity is unavailable, the decision must address that larger exposure explicitly. Don't claim a narrow permission when the underlying tool actually permits broad administration.

Reauthorize the integration without reviving old authority

A supplier can replace its compromised infrastructure while the customer's authorization still grants yesterday's capabilities. I would treat reconnection as a reviewed change to that authorization, not simply the removal of a temporary network block. Identify what the supplier may newly obtain, what it already holds, and what those capabilities can do at the resource. Those are separate questions even when one administrator manages all three.

AWS cross-account roles provide a concrete example. The AssumeRole documentation distinguishes the trust policy, which specifies who may assume the role, from the permissions policy, which specifies what the role can do. The operation issues temporary credentials; optional session policies can further limit the role's identity-based permissions (Amazon Web Services [AWS], n.d.-b). A newly issued session is therefore evidence of successful credential issuance under a configuration, not proof that the configuration matches the limited support task your incident commander approved.

For the proposed reconnection, compare the current trust relationship, role permissions and relevant resource policies with the approved capability. Do that comparison against the actual deployed state. A saved configuration file might describe what the team intended before an emergency exception or manual change. Record the approved principal, resource set and operations alongside the observed policies, and investigate unexplained differences before restoring the path. Reusing a familiar role name should not bypass that review.

The treatment of existing sessions needs its own evidence. AWS documents revocation of older role sessions through an attached deny policy, AWSRevokeOlderSessions, based on when credentials were issued. The documented cutoff does not deny a session assumed after that cutoff, and the action affects users of the specified role, not just one suspect console session (AWS, n.d.-d). This is a role-specific mechanism with documented exclusions, including service-linked roles and separate handling for IAM Identity Center permission-set sessions. It is not a universal revocation operation for every AWS identity.

The operational implication is that revoking yesterday's sessions does not settle who can obtain tomorrow's. If an untrusted supplier principal can still obtain a new session, repeating the older-session action may leave a continuing issuance path. Conversely, restricting future issuance does not itself supply evidence about each previously issued capability. The customer should separately establish the intended issuance condition and the disposition of older sessions, with tests appropriate to the actual platform and credential type.

Do not mistake an external ID for a replacement secret. AWS's third-party access guidance describes the external ID as a confused-deputy protection and explicitly states that AWS does not treat it as secret (AWS, n.d.-a). It helps bind the provider's assumption of a customer role to the intended customer context. My inference is narrower than a recovery guarantee: changing that value alone cannot establish that the provider's principal, systems or already issued credentials are trustworthy. Incident remediation must address the authority that actually creates and uses the session.

There is another subtle boundary around session policies. AWS documents that a resource-based policy naming the session ARN as principal can grant permissions that are not limited by the session policy (AWS, n.d.-c, “Session policies”). Consequently, reviewing only the policy passed when requesting a session may miss relevant resource-side authority. This is not a claim that explicit denies can be bypassed. It is a reason to evaluate the applicable policy set and test the intended resource boundary rather than equate one narrow JSON document with narrow effective access.

I would keep a reconnection record that ties the approval to a specific configuration state: approved provider principal, customer role or equivalent identity, relevant policy versions, treatment of prior sessions, permitted operation and verification result. Describe the evidence without recording temporary credential values. If the configuration changes before the approved task runs, assess whether the change affects the approval. A record that says only “vendor access restored” cannot support that comparison or explain which authority was examined.

Consider a fictional continuation of Meridian's response. The supplier requests a limited diagnostic read, and the customer approves one resource and one operation. A fresh session successfully performs that read. During review, the customer also finds a resource policy granting an additional capability to a session principal. The successful diagnostic result does not demonstrate that the extra capability is unavailable. Meridian holds broader reconnection until the effective policy is understood and the approved boundary is demonstrated. This is an analytical scenario, not an observed AWS deployment or a suggestion to probe a real supplier without authorization.

The acceptance exercise should include both the permitted operation and a deliberately selected forbidden operation using approved test resources. It should also distinguish a representative old-session case from a fresh-session case where the platform and incident constraints allow that comparison. If an old credential cannot safely be retained or reproduced, record the alternative server-side evidence and remaining uncertainty. Do not preserve exposed secrets casually to make a demonstration easier. The exercise is intended to expose a mistaken authorization assumption, not create a new credential-handling problem.

A practical objection is that a small customer may lack access to every provider-side detail. Customer-controlled policy and resource evidence still have value, but they cannot prove the security of the supplier's entire environment. Limit the resumed capability to what the combined customer evidence and scoped supplier assurance support. Where the required granularity is unavailable, escalate the larger exposure honestly; do not describe a broad restored role as a narrow permission simply because the approval ticket named a narrow task.

Make new evidence capable of reversing the decision

A reconnection decision should record its assumptions in terms that can be checked. For a limited support task, an assumption might be that the newly authorized identity is the only enabled path to the specified resource. If another standing grant is discovered, that assumption has failed even if the permitted task completed successfully. The response should revisit the grant inventory and restriction, not merely append the discovery to a closed ticket.

I would also distinguish permission expiry from the end of exposure. An authorization record can expire while an application session, downloaded dataset or queued operation remains. Specify the technical withdrawal action and how its effect will be checked. Where data has already left the customer, withdrawal cannot undo that transfer; it can stop further transfers while the confidentiality assessment continues. This is why the four channels cannot share a single closure checkbox.

Choose review triggers from the decision's dependencies: changed incident scope, failed monitoring, unexpected activity, a missed supplier evidence commitment or a material change in business tolerance. A calendar reminder alone may miss an earlier change. Assign a person able to invoke the restriction and an available escalation route if withdrawal itself would cause consequential harm. Conditional acceptance is useful only when the organization can recognize and act on the conditions it wrote down.

Reconcile the boundary before replaying work

Resumption creates a second decision beyond security access: which system's account of business activity should be trusted. During the interruption, one side may have accepted work that the other never acknowledged. A manual record may overlap an automated queue. A retry can be valid at the network layer while repeating an already completed business action.

Before bulk replay, establish the last mutually reconciled state, the disposition of in-flight items and the identifiers available to distinguish completed, pending and rejected operations. Confirm how the application handles duplicates and partial completion. This is a proposed operational acceptance practice; no cited standard supplies a universal reconciliation algorithm for every supplier service. Transaction semantics must come from the actual application and business process.

Test a bounded set of representative cases under authorization, including a successful item, a rejected item and a duplicate or interrupted item where relevant. Define what evidence permits increasing volume and which discrepancy requires stopping. “The connection works” establishes much less than “the process produces correct, attributable outcomes.” Monitoring should include business exceptions as well as security events.

Restoration of throughput must not erase the incident record. Preserve what was withheld, manually handled, corrected or deliberately abandoned, with accountable decisions. If an export cannot be reconciled, distinguish availability from accuracy. A service can be reachable and still be unsuitable for consequential processing. That distinction belongs in the executive update so that a technical restoration announcement does not close unfinished business recovery.

A fictional example: one supplier, three different decisions

Consider fictional Meridian, which uses a supplier for invoice processing, remote application support and software updates. The supplier reports ransomware affecting its operations, with identity scope still under investigation. Meridian has no established evidence of unauthorized activity in its own systems. Finance has a manual approval process, but it cannot assume that supplier-side work accepted before the outage was never completed.

Meridian restricts the delegated support path, pauses the update channel and preserves relevant local logs and transaction states. Its incident and operational owners assess the interruption together. Finance records manual approvals in a controlled queue and holds automated replay. The supplier is asked for distinct information about tenant data, support identities and releases, rather than a single date on which everything will be safe again.

Later, the supplier can provide a bounded export with authenticated delivery and records sufficient for Meridian to compare it with its own ledger. That evidence may support receiving and reconciling the export under defined controls. It does not resolve the earlier uncertainty about standing support privileges. Meridian keeps those blocked pending identity-specific assessment and retains the hold on updates while release provenance remains unresolved.

Suppose the comparison reveals an invoice already processed by the supplier but recorded locally as awaiting completion. Meridian resolves that item before replaying the queue. The case illustrates two separate acceptance failures that a general restoration announcement would miss: excessive authority could return without being assessed, and valid business work could be duplicated without any new malware. No numerical probability or claimed client outcome is attached to the example.

Objections, limits and the decision leaders actually own

One objection is that this approach asks too much during a fast-moving incident. That is fair if the team turns every channel into an elaborate questionnaire. Start with the paths that can cause consequential harm and the next operating decision. Record enough to support that decision, then deepen the assessment as evidence arrives. Brevity should reduce administration, not obscure which capability remains exposed.

Another objection is that a customer cannot investigate a supplier's entire environment. It usually cannot establish everything from its own boundary, and the framework doesn't require that impossible claim. It requires identifying what the available evidence supports, what depends on supplier assurance and what remains unresolved. A consequential unresolved dependency may require continued restriction, an alternative service or explicit executive acceptance within the organization's actual authority.

The framework also cannot predict attacker persistence, prove the completeness of supplier disclosures or remove the cost of interruption. The sixteen sources provide technical and governance foundations, not an experimental comparison showing that this method outperforms alternatives. Old forensic and contingency guidance is used for stable concepts, not contemporary platform commands or legal duties. Sector-specific safety, privacy and contractual decisions require the relevant specialists and facts.

Preparation reduces the amount that must be invented during the event. CSF 2.0 explicitly includes relevant suppliers and third parties in incident planning, response and recovery (NIST, 2024, GV.SC-08). Before the next event, agree contacts, customer-access boundaries, evidence-sharing arrangements and exercise conditions. Then test whether the organization can restrict and resume the actual dependency, including the business reconciliation that a connectivity test misses.

For the earlier purchasing decision, see what security evidence to require before approving a critical supplier. For shared technical response responsibilities, see cloud incident ownership. For rebuilding your own environment, see how to prove a ransomware restore is ready for production. These decisions connect, but their evidence and acceptance conditions differ.

If a supplier incident has left your leadership team choosing between continued disruption and poorly understood access, contact me about a supplier incident readiness or reconnection assessment. We can define the critical dependencies, the evidence each operating decision needs and a controlled path from restricted service to verified business operation. The useful deliverable is a decision your team can execute and review, with the remaining uncertainty visible.

References