APT10 made the provider’s admin account the client’s perimeter

The December 2018 APT10 indictment described intrusions that reached clients through their managed service providers’ own credentials and connections. A client’s security perimeter therefore included its provider’s administrative environment, and contracts had to say so.

Looking Back revisits past weeks one at a time, up to the most recent, and asks what it meant for the banks, health systems and public agencies we serve. The series was written in 2026, with the benefit of hindsight, and each article says which week it looks back on.

Two-minute summaryThe dek and the first sentence of each section, quoted from the article

The December 2018 APT10 indictment described intrusions that reached clients through their managed service providers’ own credentials and connections. A client’s security perimeter therefore included its provider’s administrative environment, and contracts had to say so.

  1. The week On December 20, 2018, the Justice Department announced the unsealing of an indictment in Manhattan federal court against two nationals of the People’s Republic of China.
  2. How a provider’s access became an attacker’s The Justice Department’s account of the method is short, and each step depends on a design choice the provider or the client made.
  3. The credentials that could act in your environment TA17-117A described compromises involving system administrators’ credentials used to access trusted domains, and the malicious use of certificates.
  4. The connections between provider and client TA18-276B gave specific architectural advice to provider customers.
  5. Who could see activity across the boundary The alleged method was designed to blend in.
  6. The contract that allocates responsibility A provider contract written around service levels, such as response times, uptime and ticket resolution, can describe the provider’s security only in general terms.
  7. The strongest objections One objection is that APT10 was an intelligence operation aimed at intellectual property, and a community bank or a county government had nothing it wanted.
  8. How it differs for banks, health systems and agencies Banking organizations supervised by the federal banking agencies operate under third-party risk management expectations.
  9. With hindsight The government kept returning to the same mechanism.

The week

On December 20, 2018, the Justice Department announced the unsealing of an indictment in Manhattan federal court against two nationals of the People’s Republic of China. They were charged with conspiracy to commit computer intrusions, conspiracy to commit wire fraud and aggravated identity theft. The department said both were members of a group known in the security community as Advanced Persistent Threat 10, or APT10. It said they worked for a company in Tianjin and acted in association with the Tianjin State Security Bureau of China’s Ministry of State Security.

The indictment, as the department described it, alleged two campaigns. The first, from about 2006, targeted more than 45 technology companies and U.S. government agencies in at least 12 states. The second, beginning at least in about 2014, targeted managed service providers. The department defined these as companies that remotely manage the information technology infrastructure of businesses and governments. The group was alleged to have targeted providers in order to use their networks to reach their clients’ networks.

The department laid out the method in three steps. First, the group installed malware on provider computers, disguised as legitimate operating system files to avoid antivirus detection. Second, it used stolen administrative credentials to connect to other systems in a provider’s network and its clients’ networks, reaching computers that were not yet infected. Third, it packaged data in encrypted archives and used stolen credentials to move one client’s data to other compromised computers of the provider or its other clients before taking it out.

The alleged victims of the provider campaign were in at least 12 countries. They included, according to the department, a global financial institution, a healthcare company, a biotechnology company, telecommunications and manufacturing companies, and others. The department said one provider with offices in the Southern District of New York had been compromised, along with data of certain of its clients in sectors including banking and finance and healthcare. Separately, the group was alleged to have compromised more than 40 computers to steal data on more than 100,000 Navy personnel, including Social Security numbers.

The same day, the government’s cyber incident center updated an alert first published in April 2017, Alert TA17-117A, to reflect the public attribution of the activity to the Chinese government. The alert had been issued by the National Cybersecurity and Communications Integration Center, which by then was part of the Cybersecurity and Infrastructure Security Agency. It described a campaign since at least May 2016 that used stolen administrative credentials and certificates and placed malware implants on critical systems. It said some victims were IT service providers, whose compromised credentials could be used to reach customer environments.

The alert explained why. Many service providers, it said, use common core infrastructure that should be logically isolated to support multiple clients. An intrusion into that infrastructure lets an adversary use stolen credentials to reach customer environments. A companion alert from October 2018, TA18-276B, had already said that providers generally have direct and unfettered access to their customers’ networks. It said using one significantly increases an organization’s infrastructure footprint and its number of privileged accounts.

The charges were allegations. The department’s release said the defendants were presumed innocent unless proven guilty.

The APT10 indictment was read at the time mainly as a statement about China. The government named a state-linked group and charged two of its alleged members. For the technology and security leaders of banks, health systems and agencies that used managed service providers, it carried a different message. The indictment and the alert that accompanied it described, in official language, how a provider’s access became an attacker’s access.

Our argument is that the case showed a client’s security perimeter included its provider’s administrative accounts and the infrastructure those accounts ran on. The allegations did not describe a client breach that happened to involve a vendor. They described a client environment entered through the provider’s own credentials, over the provider’s own connections, in a way that looked legitimate to the client’s monitoring. A perimeter drawn around the client’s own network leaves that path outside it. The case showed why that line is in the wrong place.

We work through four parts of that perimeter, after a short account of the mechanism: the credentials that could act in a client’s environment, the connections that linked provider and client, the visibility that either side had into activity across them, and the contract that allocated responsibility for all three. A sixth section tests the argument against the strongest objections. The last looks at how it plays out for banks, health systems and public agencies.

The sector is cross-sector because the alleged victims spanned finance, healthcare and government, and the alerts addressed every organization that used a provider. The analysis is not legal advice. Each organization’s obligations depend on the rules that apply to it, which it should confirm with counsel.

How a provider’s access became an attacker’s

The three steps, read as architecture

The Justice Department’s account of the method is short, and each step depends on a design choice the provider or the client made. In the first step, malware was installed on provider computers and disguised as operating system files. That required a foothold inside the provider, on machines that administered clients.

In the second step, stolen administrative credentials were used to connect to other systems in the provider’s network and in clients’ networks. That required credentials that worked in more than one place, and network paths that let a provider machine reach client machines. In the third step, one client’s data was moved through other compromised computers of the provider or of other clients before it left. That required paths between clients as well as between each client and the provider.

None of those steps involved breaking a client’s external defenses. Each used access the client had granted, or access the provider had built for its own convenience.

Why legitimate access hides an intrusion

Alert TA17-117A said user impersonation through compromised credentials was the adversary’s primary mechanism. It said the actors could gain full access to networks and data in a way that appeared legitimate to existing monitoring tools. TA18-276B added that the actors used living off the land techniques: legitimate credentials, trusted off-the-shelf applications and system tools already present on client networks.

The October alert named examples. It said PowerSploit, a set of PowerShell and Visual Basic scripts originally built for penetration testing, was widely misused. It said the actors had been seen using Robocopy, a Microsoft command line tool, to move stolen data back through provider environments. Those tools are what administrators use every day. A client that watched for unusual software would see nothing unusual.

How long it went on

The timeline in the official record matters as much as the method. The Justice Department said the provider campaign began at least in about 2014. TA17-117A described a campaign occurring since at least May 2016 and was first published in April 2017. The public attribution came in December 2018. Whatever the exact dwell time in any one network, the record describes years of activity, not days.

That has a direct consequence for clients. An investigation can only look back as far as the logs reach. TA17-117A recommended at least 180 days of netflow and a year of system logs for mission critical systems. A client that kept only a few weeks of logs could not have answered whether provider credentials had been misused in its environment over the period the indictment described. It could only say what it saw recently.

What a client could not have known

A client reading the indictment could not tell whether its own provider was among those compromised. The Justice Department described victims by sector and country, not by name, apart from government bodies. That uncertainty is the normal condition for clients after a provider-focused campaign is disclosed. The response cannot wait for confirmation. It has to start from the assumption that provider credentials might have been exposed, and work through the accounts, connections and logs that would show misuse.

Why a provider is an efficient target

TA18-276B explained the economics plainly. Providers serve many customers, which gives them economies of scale. It also means a compromise in one part of a provider’s network can spread to other customers. For an attacker, one provider can open many client doors. That changes the threat model for each client, because the client’s exposure depends on how the provider defends every other client as well as itself.

The credentials that could act in your environment

What the alerts said about administrative accounts

TA17-117A described compromises involving system administrators’ credentials used to access trusted domains, and the malicious use of certificates. Its recommendations addressed accounts directly. It advised reducing the number of domain and enterprise administrator accounts and giving privileged users separate non-privileged accounts for everyday work. It advised two-factor authentication, with soft tokens kept off the device requesting access. It advised tracking privileged roles, putting privilege changes under change control and alerting on them.

Those are standard controls for an organization’s own staff. The case made the point that they had to apply equally to accounts held by the provider’s staff. A provider technician with domain administrator rights in a client’s environment held the same power as the client’s own administrator. If the technician’s credentials were stolen in the provider’s network, the client’s strongest account was in an attacker’s hands.

Accounts that span clients

The alleged method relied on credentials that worked across systems, and the data movement step relied on paths across clients. A provider that used the same administrative credentials for many clients turned one theft into many. The 2022 joint advisory discussed in the hindsight section made the point explicitly, telling providers not to reuse administrative credentials across customers. The risk was visible in the 2018 material: shared core infrastructure, administrative credentials and bidirectional movement.

Certificates and service accounts

TA17-117A also mentioned the malicious use of certificates. Certificates authenticate machines and connections, and they can live longer and get less attention than passwords. A provider connection authenticated by a certificate that never changes gives a long-lived key to whoever copies it. TA18-276B’s advice to update connection certificates at least annually, and consider rotating them every six months, addressed exactly that.

Service accounts raise a similar issue. Remote management agents and backup tools can run under accounts with broad rights and no human owner. They are not covered by the joiner and leaver processes that catch staff accounts. A client inventory that lists only named provider staff misses them. The inventory should list every non-human identity the provider uses in the client’s environment, with its rights and the date its secret last changed.

What a client could do in 2018

A client did not need a new framework to act on this. It needed an inventory and a few decisions.

  • List every provider account with any privilege in the client’s environment, including service accounts and accounts used by automated tools.
  • Name a client-side owner for each, who approves its scope and reviews it on a schedule.
  • Reduce standing privilege, so that provider staff hold high privilege only for defined tasks and periods.
  • Require strong authentication on every provider account, enforced in the client’s own identity system where possible.
  • Disable accounts promptly when provider staff change roles or the contract ends.

Each step is ordinary identity hygiene. The difference is that it extends to people the client does not employ, using systems the client may not control.

Who owns the identity

A harder question is whose directory holds the provider’s accounts. If the provider’s staff log in with accounts in the client’s directory, the client can enforce its own policies and see each login. If they log in through the provider’s own directory, trusted by the client, the client depends on the provider’s controls. TA17-117A’s reference to access through trusted domains points at the second arrangement. Trust relationships between directories are convenient. They also carry a compromise across the boundary.

A client can review each trust it has granted. It can record why the trust exists, which accounts on the other side can use it, and what they can reach. Where a trust can be replaced by named accounts in the client’s own directory, the client gains visibility and control at the cost of some administrative work for the provider. Where it cannot, the client can narrow the trust to the specific systems under management and log every use of it.

The connections between provider and client

TA18-276B gave specific architectural advice to provider customers. It recommended a dedicated virtual private network for the provider connection, using certificate-based authentication and hosted on its own device. It said the connection should terminate in a demilitarized zone isolated from the internal network. It said traffic should be confined to the networks and protocols needed for the service, with everything else blocked and failed attempts logged.

It also recommended updating the connection’s authentication certificates at least annually, and suggested considering rotation every six months. It said all connection attempts should be logged centrally and reviewed. It advised separating internal networks by function, location and risk profile, with access control lists governing traffic between them.

Why that design limits the damage

A dedicated, restricted connection does not stop an attacker who holds valid provider credentials. It narrows what those credentials can reach. If the provider’s connection lands in an isolated zone and can reach only the systems under management, a compromise of the provider reaches only those systems. If it lands on the internal network with broad routing, a compromise reaches everything.

The same logic applies inside the provider. TA17-117A said shared core infrastructure should be logically isolated between clients. The third step of the alleged method, moving one client’s data through another client’s compromised computers, shows what happens when that isolation is weak. A client cannot build the provider’s internal segmentation. It can ask whether it exists and require evidence.

Paths between clients

The third step of the alleged method deserves separate attention. Data from one client was moved to other compromised computers of the provider or of its other clients before leaving. That means the attacker could reach from one client environment to another through the provider. A client whose own environment is well defended can still be used as a staging point if the provider’s network connects it to weaker ones.

For a client, that is both a confidentiality risk and a reputational one. Its systems could hold another organization’s stolen data. Its outbound traffic could carry that data to an attacker. A client cannot see whether the provider permits paths between clients. It can require that the provider does not, and it can watch its own connection for traffic that does not match the services it buys.

The trade-off of isolation

Isolation costs the provider efficiency. Shared tooling, shared jump hosts and shared administrative accounts are part of what makes a provider cheaper than an in-house team. Asking a provider to run a separate management environment for each client may price the service out of reach for small clients. A reasonable middle ground is isolation at the points that matter most: separate credentials per client, separate network paths into each client, and controls that stop one client’s connection from reaching another’s.

Remote management tools

Providers manage clients through remote management and monitoring software. Those tools are designed to run commands and move files across many machines at once. An attacker with access to them has the same reach. The 2018 alerts did not single out a product, and neither do we. The architectural point is that the management plane is part of the client’s attack surface and needs the same scrutiny as any other path into the network.

What a client would map

A useful exercise is a connection map for each provider. It shows every path by which the provider can reach the client: network tunnels, remote management agents, cloud console access, directory trusts and shared file locations. For each path, it shows what systems are reachable, which accounts can use it, and where the logs go. Drawing the map can surface paths nobody remembers approving.

Who could see activity across the boundary

The visibility gap

The alleged method was designed to blend in. Valid credentials, ordinary tools and traffic routed through the provider meant that each side saw only part of the picture. The client saw a provider account doing provider work. The provider, if it watched its own network at all, saw its own machines talking to clients. The movement of one client’s data through another client’s computers could be invisible to both.

TA17-117A said that, in some cases, the malware was found only in memory, with no evidence on disk. That makes detection depend on monitoring that captures activity as it happens, not on forensic examination after the fact.

What the alerts asked for

TA17-117A’s network recommendations were specific. It said to capture netflow at every ingress and egress point of network segments, including internal ones, with a minimum retention of 180 days. It said to retain packet capture for at least 24 hours. It recommended storing system logs of mission critical systems for at least a year in a security information and event management system. It recommended reviewing application log fields to make sure they would help an investigation.

TA18-276B said organizations should configure logs to detect incidents and identify their type and scope. It said clients that do not conduct most of their own network defense should work with their provider to determine what security they can expect.

Whose logs, and who reads them

Many clients outsource monitoring to the same provider that manages their systems. That arrangement concentrates both the access and the oversight of the access in one firm. If the provider’s environment is compromised, the attacker may be able to see, or affect, the monitoring meant to catch it.

A client can reduce that risk without taking monitoring in-house. It can require that logs of provider activity in its environment flow to a store the provider cannot alter. It can arrange for a second party, or its own small team, to review provider account activity against expected work. It can ask for notice of any security event in the provider’s own environment that could affect it, even if the provider has not found evidence of client impact.

Responding across the boundary

Detection is half the problem. When a client suspects that provider credentials were misused, the response involves two organizations with different priorities. The provider wants to understand exposure across all its clients and may not share details about other clients. The client wants to know what happened in its own environment, and quickly.

TA18-276B’s response advice applies across that boundary. It recommended an incident response plan, written guidelines that prioritize incidents by mission impact, and procedures and out-of-band lines of communication for incident reporting with internal and external parties. It recommended exercising response to different intrusion scenarios regularly.

For a client, that means the incident plan names the provider’s security contact and an out-of-band way to reach them. It says who decides to cut the provider connection, and what the client does to keep operating when it does. It says how evidence in the provider’s environment will be preserved and shared. An exercise that includes the provider, run once a year, tests whether any of it works.

What the provider should report

Notice terms can be written around confirmed breaches of client data. The APT10 mechanism shows why that trigger comes too late. By the time a provider confirms that a particular client’s data left, the attacker may have held the provider’s administrative credentials for a long time. The useful trigger is earlier: a suspected or confirmed compromise of any system or account that can reach the client.

A client can define the events it wants to hear about. They include compromise of provider administrative accounts, compromise of the provider’s remote management tooling, malware found on provider machines that administer clients, and law enforcement or government notification to the provider about its environment. Each of those is a signal the client needs in order to decide whether to cut the connection and look for misuse.

Measures a client could track

Visibility can be measured. Useful measures include the share of provider accounts whose activity reaches the client’s log store, the time between a provider staff change and the related account change, the number of provider connections reviewed in the period, and the number of provider logins outside expected hours or tasks. None needs new technology. Each needs someone to own it.

The contract that allocates responsibility

What a service-level contract leaves out

A provider contract written around service levels, such as response times, uptime and ticket resolution, can describe the provider’s security only in general terms. Such a contract can say little about how the provider protected its own administrative environment, whether it reused credentials across clients, or what it would tell a client about an intrusion in its own network.

TA18-276B named the fix in general terms. It said clients should understand the supply chain risk of their provider and manage it across security, legal and procurement groups. That instruction puts the contract at the center. Security can specify what is needed. Only procurement and legal can make it binding.

Terms that follow from the mechanism

Each part of the perimeter described above translates into a contract term:

  1. Credentials. No reuse of administrative credentials across clients, strong authentication on every account with access, and client approval of privileged accounts.
  2. Connections. A defined, restricted connection method, with no other paths, and notice before any new path is added.
  3. Isolation. Logical isolation of the client’s data and management systems from other clients, with evidence on request.
  4. Visibility. Delivery of logs of provider activity in the client environment, and notice of security events in the provider’s own environment that could affect the client.
  5. Exit. Removal of all provider accounts and connections at termination, confirmed in writing.

Each of these terms is familiar on its own. Together they shift the contract from describing a service to describing a boundary.

Evidence a client can ask for

A term is only as good as the client’s ability to check it. A client can request the provider’s list of accounts with access to its environment and compare it with its own directory. It can request third-party assurance reports and read the sections on logical access and segregation. It can test the connection rules itself, by trying to reach systems that should be out of scope. It can ask, at renewal, how the provider responded to the 2018 alerts. A provider that can answer each of these requests with documents, rather than assurances, has already done much of the work the contract asks for.

The provider’s own suppliers

Providers rely on suppliers too: software for remote management, hosting for their tools, and sometimes subcontracted staff. Each is another party with a path toward the client. TA18-276B advised clients to understand the supply chain risk associated with their provider. A contract can ask the provider to disclose material subcontractors that will have access to the client’s environment, to bind them to the same security terms, and to give notice before adding new ones.

Renewal as the lever

Most clients cannot reopen a contract mid-term. They can use renewal. A client that prepares a short security schedule, based on the five terms above, can present it at every renewal and ask the provider to mark which terms it already meets. The gaps become the negotiation. Over a few renewal cycles, the schedule becomes standard, and the client’s control over the boundary grows without a single dramatic renegotiation.

The cost and who bears it

Every one of these terms costs money. Dedicated connections, separate credentials per client and log delivery add work for the provider, which will price it. Smaller clients have less bargaining power to demand bespoke terms. The trade-off is real. The alternative, though, is accepting a perimeter set by the provider’s cheapest design.

The strongest objections

Nation-state espionage is not a typical client’s threat

One objection is that APT10 was an intelligence operation aimed at intellectual property, and a community bank or a county government had nothing it wanted. The premise is partly right about motive. It is wrong about method. The technique described, stolen provider credentials used to reach clients, does not depend on who uses it. TA18-276B noted that a provider footprint creates a larger attack surface for cyber criminals and nation-state actors alike.

Providers are better at security than clients

A second objection is that providers usually defend systems better than their smaller clients could. That may be true for a given provider. The argument here is not that clients should leave providers. It is that the provider’s administrative environment is part of the client’s perimeter, so its quality matters to the client and should be checked like any other part. A capable provider will welcome that, and should be able to evidence it.

Contract terms do not stop attackers

A third objection is that contracts are paper and intrusions are technical. A contract does not stop an intrusion. It does decide whether the controls that would limit one exist, whether the client can see their state, and whether the client learns of a compromise in time to act. The terms above are technical controls with a signature attached.

The inventory is too much for small organizations

A fourth objection is practical. A small credit union or a rural clinic may have one provider, no security staff and no capacity to map connections or review logs. That is true, and it argues for simplicity rather than inaction. The smallest useful version is short: a list of provider accounts, strong authentication on all of them, one restricted connection, logs kept somewhere the provider cannot change, and a contract clause requiring notice of compromise. A small organization can ask a peer, an auditor or its association for help with the rest.

Attribution changes nothing for defenders

A fifth objection is that naming a state and indicting two people did nothing for a client’s defenses. The technique was already described in the 2017 and October 2018 alerts. The objection is fair about the indictment’s direct effect. Its value to clients was different. It put the mechanism on the public record with the weight of a federal case, in a form a board or an audit committee would read. For security leaders who had struggled to get provider controls funded, that record was useful evidence.

Where the argument is weaker

The argument is weakest for providers that never hold privileged access, such as a vendor that supplies software the client installs and runs itself. It is strongest for full-service providers that administer directories, servers and networks. Clients should spend their effort accordingly, starting with providers that hold the most privilege.

How it differs for banks, health systems and agencies

Banks and credit unions

Banking organizations supervised by the federal banking agencies operate under third-party risk management expectations. In June 2023 the Federal Reserve, the FDIC and the OCC issued joint final guidance that replaced each agency’s earlier general guidance. It said using third parties does not diminish a banking organization’s responsibility to operate safely and soundly and in compliance with law. It listed information security implications, including a third party’s access to the bank’s systems and confidential information, among the matters to consider in planning.

For a bank, the APT10 mechanism maps onto topics that guidance covered: the right to audit and subcontracting among contract provisions, and incident reporting in due diligence. On incident reporting, it described the value of reviewing whether a third party has clearly documented processes, timelines and accountability for identifying, reporting, investigating and escalating incidents. A bank whose core or network provider holds administrative access should be able to show its examiner how that access is controlled and monitored, and what the provider must report. Credit unions are supervised by a different agency. The 2026 proposal discussed in the hindsight section included the National Credit Union Administration among its issuers.

Health systems and ambulatory networks

Under the HIPAA Security Rule, a provider that creates, receives, maintains or transmits electronic protected health information for a covered entity is a business associate. The required business associate contract must provide that the business associate will comply with the applicable security requirements, report security incidents it becomes aware of, and bind its own subcontractors to the same terms.

The incident reporting term is where the APT10 mechanism bites. A provider that learns of a compromise in its own administrative environment has learned of something that may be a security incident affecting the client. The contract should make clear that such events are reported, and how fast. Ambulatory networks, which may rely entirely on a provider, have the most to gain from that clarity and the least ability to monitor independently.

Public agencies

Federal civilian agencies, states and localities face different rules, but the same architecture. TA17-117A’s recommendations were written partly for federal networks, referring to Trusted Internet Connection locations and to Personal Identity Verification authentication to control access to VPN equipment. A state or county agency that uses a provider for its network and directory faces the same exposure as a bank, often with smaller security staff and fixed procurement terms.

Organizations that depend on one provider

Across all three sectors, the hardest case is the organization that has handed its whole technology function to one provider. It has no internal administrator who could notice unusual provider activity. Its monitoring, if any, is run by the same provider. Its contract was often signed on price.

For that organization, the most valuable step is independence at one point. That could be an outside party that reviews provider account activity each quarter. It could be a log store under the organization’s own account. It could be an annual test of whether the provider’s connection reaches only what it should. One independent check is far better than none, and it is affordable at small scale.

Procurement as the control point

For agencies, the contract is usually the procurement vehicle, and its terms are set at solicitation. Changing terms after award can require a contract modification, with its own cost and delay. The practical step is to put the provider security terms into standard solicitation language, so that each new award includes them by default rather than by negotiation.

Agencies also face a staffing question. A provider contract often exists because the agency lacks the staff to run the systems itself. The same shortage limits its ability to oversee the provider. Agencies can share the oversight burden: a state can write model provider terms for its counties, or a shared service can review provider access logs for several small agencies. The oversight function is easier to share than the systems themselves.

With hindsight

The government kept returning to the same mechanism. CISA’s page for alert TA18-276B, on providers, shows it last revised June 30, 2020, though its revision history lists only the October 2018 original. On May 11, 2022, the cybersecurity authorities of the United Kingdom, Australia, Canada, New Zealand and the United States, including CISA, the NSA and the FBI, issued a joint advisory on cyber threats to managed service providers and their customers. It said they expected malicious actors, including state-sponsored groups, to increase their targeting of providers to exploit trust relationships with customers.

The 2022 advisory turned the 2018 lessons into contract language. Its first tactical actions were to disable unused accounts, enforce multifactor authentication on provider accounts that access customer environments, and make contracts identify who owns each security role. It told customers to require that providers not reuse administrative credentials across customers. It said providers should segregate customer data sets and services from each other. It said customers should require visibility of the provider’s presence, activities and connections in their networks, and notice of security events on the provider’s infrastructure. It warned that disabling provider accounts can be overlooked when a contract ends. It said providers should treat all accounts with access to customer environments as privileged and protect them with multifactor authentication. It recommended that all organizations keep their most important logs for at least six months, because incidents can take months to detect. It added that customers should verify their contracts include security measures in line with their own requirements.

Bank supervisors addressed providers through general third-party guidance. The June 2023 interagency guidance on third-party relationships discussed information security access, the right to audit, subcontracting and incident reporting. In November 2025, the OCC sought information on community banks’ engagement with core service providers and other essential providers. On September 15, 2026, the OCC, the Federal Reserve, the FDIC and the NCUA proposed new third-party risk management guidance that would rescind and replace the existing guidance. As checked on November 14, 2026, it was a proposal, with comments due by November 16, 2026.

The 2023 guidance was directed to all banking organizations the three agencies supervised. It replaced each agency’s existing general guidance on the topic. The agencies said they had received 82 comment letters on the 2021 proposal that preceded it. The 2026 proposal said it would help banking organizations tailor third-party risk management to their size, complexity and risk profile, and to the nature of each relationship. The OCC’s 2025 request for information had asked for comments by January 27, 2026.

Neither the 2023 guidance nor the 2026 proposal named managed service providers as a separate category. Both treated them as third-party relationships, to be managed according to the risk they posed.

The HIPAA Security Rule’s business associate provisions, as in effect on November 1, 2026, still required incident reporting by business associates and the flow-down of terms to subcontractors.

What held up was the 2018 reading of the threat. The provider’s administrative environment stayed part of the client’s perimeter, and the 2022 joint advisory said so in contract terms. Whether those terms reached contracts is harder to show from the public record. The 2022 advisory still had to tell customers to ask for them.

Spotted an error? Tell us and we will check it and correct the article where needed.

  • APT10
  • managed service providers
  • privileged access
  • third-party risk
  • CISA