Skip to content
Knowledge Base

Third-party vendor access to critical systems: how do you control it under DORA?

The integrator of your core system, the firm that maintains your databases, the infrastructure service provider — each of them logs into systems your institution is accountable for, not theirs. We show how to turn DORA and Delegated Regulation 2024/1774 into concrete decisions about the account, the session and the evidence for audit.

The integrator of your transaction system, the firm that maintains your databases, the storage and network service provider — each of them logs into systems that your institution, not they, answers for before the supervisor. Third-party vendor access to critical systems usually originates in a contract: an account for the project kick-off, a VPN for the whole team, a password in a message from the project manager. Then nobody removes it, because nobody formally owns it.

DORA does not allow responsibility to be shifted to the vendor, and Delegated Regulation (EU) 2024/1774 goes down to the account level: a unique identity for every technician, privileged access only when needed, a rights review every six months. We cover contracts with ICT providers and the register of information in our overview of the DORA regulation. Here we deal with what a contract cannot demonstrate: who exactly logs in, for how long, with whose approval and what remains as evidence.

Why is a vendor with administrator rights a risk that no contract closes?

Because the contract governs the relationship with a company, while the system sees an account — and it is the account, not a clause, that decides what happens in the database at two in the morning. A contract will oblige the vendor to follow your policies and report incidents, but it will not tell you which of its twenty engineers logged into the settlement server yesterday and why.

In practice, vendor access has three typical weaknesses. A shared account — a single “vendor_admin” for the whole support team, with a password known to everyone who has ever worked on the project. Standing access — a VPN active for the entire term of the contract, even though actual work takes a few days a quarter. No owner — nobody on the institution’s side is responsible for the account, so nobody notices that the technician left the vendor six months ago.

Article 28(1)(a) of DORA settles the question of responsibility: a financial entity using ICT services under a contract remains fully responsible at all times for meeting its obligations under the regulation. An incident from a vendor account is therefore the institution’s incident — in reporting, in the supervisor’s assessment and in the conversation with the management board.

What does DORA require from controlling vendor access to ICT systems?

Regulation (EU) 2022/2554 itself sets the principle, and Delegated Regulation 2024/1774 (the regulatory technical standards for the ICT risk management framework) turns it into a checklist. Article 9(4)(c) of DORA requires policies that limit physical and logical access to ICT assets to “what is required for legitimate and approved functions and activities only”. Point (d) requires policies for strong authentication.

The delegated regulation adds details that concern vendors directly:

Provision (Delegated Regulation 2024/1774)What it means for a vendor account
Art. 20(2)(a)every staff member of an ICT third-party service provider has a unique identity and their own account
Art. 21(a)rights follow the least-privilege principle, including for remote and emergency access
Art. 21(c)as few generic and shared accounts as possible, every action attributed to a person
Art. 21(e)(ii)privileged and administrator access only when needed or ad hoc
Art. 21(e)(iii)withdrawal without undue delay once access is no longer necessary
Art. 21(e)(iv)rights review at least every 6 months for systems supporting critical or important functions
Art. 21(f)(ii)strong authentication for remote and privileged access

Article 21 adds that financial entities use dedicated accounts for administrative tasks where possible and, “where feasible and appropriate”, put automated privileged access management tooling in place. Article 19(b), in turn, requires vendor staff who use the institution’s assets to know and comply with its security policies.

In short: DORA does not require a specific product, but it does require an outcome that cannot be shown without a tool — named, time-bound, approved and logged access for every person on the vendor’s side. Scope: entities subject to the full ICT risk management framework; the simplified framework under Article 16 of DORA has separate provisions.

Which accounts and secrets does a vendor hold, and where should they be kept?

More than the contract suggests — which is why the inventory starts with systems, not with the vendor list. On servers and in applications, look for everything a vendor can authenticate with:

  • named technician accounts in operating systems, databases and application consoles,
  • technical and service accounts created during implementation (often with full rights “just in case”),
  • SSH keys and certificates uploaded to servers by the project team,
  • API keys and tokens used by the vendor’s monitoring tools or integrations,
  • accounts in cloud consoles and with managed service operators from which the vendor administers the institution’s resources.

The organising principle is simple: the vendor should know its login credentials for the access gateway, but not the passwords to target systems. Privileged credentials are kept in a password vault, which injects them into the session and changes them once it ends. The technician does not see the database administrator password, so they cannot write it down, pass it on or use it after the engagement ends.

Technical accounts that cannot be moved into the vault get an owner on the institution’s side and a review date; an account without an owner is the first to be disabled.

How do you grant a vendor access for the duration of a task, not the contract?

Through a request tied to specific work: access comes into existence when there is a ticket and expires by itself when the window closes. This is how we read Article 21(e)(ii) of the delegated regulation — privileged access on a “need-to-use or an ad-hoc basis”, not for the entire term of the contract.

A vendor request should contain four elements: the ticket or change number, the target system, the scope of work and the time window. It is approved by the system owner on the institution’s side, not by the security team alone — because the owner knows whether the settlement system can take a change on Thursday at 10 p.m. For high-risk work (changing the configuration of the core system, operations on production data) it is worth requiring two approvals, i.e. the four-eyes principle.

An emergency path has to be designed separately. An outage at three in the morning cannot wait for a manager’s signature, but it cannot bypass controls either. The “access first, justification afterwards” variant works well: the on-call engineer gets access immediately, the session is recorded and must be reviewed the next day. Article 21(a) of the delegated regulation applies the least-privilege principle to emergency access as well, so its scope is also limited to what is needed.

How do you supervise a vendor session and what should you log?

Supervision has two layers: a recording of every session and live viewing where the risk is high. Watching all sessions in real time is unrealistic; live viewing with the ability to terminate makes sense for changes to critical systems and for a new vendor’s first jobs.

The record has to reconstruct what happened, not merely confirm that someone connected. Privileged session recording covers the screen or command stream as well as metadata: who, from where, to which system, on which target account, in which window and under which request. Without that metadata the recording is a film from which an auditor can read nothing.

Article 12 of Delegated Regulation 2024/1774 describes what logging should look like:

  • events related to access control and identity management are logged (paragraph 2(c)(i)),
  • logs are protected against tampering, deletion and unauthorised access (point (d)),
  • system clocks are synchronised with a reliable time source (point (f)), so a session can be matched with the target system’s logs.

The institution sets the retention period for recordings itself, based on its ICT risk assessment — the provision gives no number. The justification should take into account the time it takes to detect an incident, the vendor’s liability period for the work performed and the fact that the recording contains the technician’s personal data.

How do you turn a session record into evidence for the auditor and the supervisor?

By linking three records that usually sit in different places: the access request, the session recording and the outcome of the rights review. The auditor does not ask whether the institution “has PAM”, but who at the vendor had access to a critical system in a given quarter, who approved it and whether the access expired.

Well-prepared evidence answers four questions without manually assembling spreadsheets:

  1. Who — a list of the vendor staff identities with the dates access was granted and withdrawn. Article 20 of the delegated regulation requires a record of all assigned identities, kept also after the contractual relationship ends.
  2. On what basis — an approved request linked to every session.
  3. What they did — the session recording and metadata, time-consistent with the target system’s logs.
  4. Whether anyone checked — the minutes of the rights review, at least every six months for critical systems, with “keep / withdraw” decisions.

The same material helps after an incident — the analysis starts from the list of vendor sessions in the event window — and in a dispute with the vendor, when a recording with an approved scope of work protects both sides.

Does a vendor have to connect over a VPN and install an agent?

No, and often it should not. A VPN into the internal network gives the vendor visibility of the whole segment even though the task concerns a single server, and an agent on its workstation is software the institution does not control.

The alternative is a session broker: the vendor connects to a gateway which, after authentication and checking the request, opens a session only to the designated system on the designated target account. From the network’s point of view, the vendor is not “inside” — it sees a single session. Some solutions go further and reverse the direction of the connection: the tunnel is set up from the institution’s side as an outbound connection, so the gateway does not accept traffic from the internet.

The VPN does not disappear from every scenario, though. A managed service provider that administers the entire network infrastructure may need broader connectivity than individual sessions. The decision is therefore made separately for each type of access, and the criterion is the scope needed for the task, in line with Article 9(4)(c) of DORA.

How do you revoke access when the engagement with a vendor ends?

With a single action in a single place — the strongest argument for a central gateway. Article 21(e)(iii) of the delegated regulation requires access rights to be withdrawn without undue delay once access is no longer necessary. With scattered access (local accounts, a separate VPN, SSH keys in many places), “without undue delay” means weeks of searching.

Revocation comes in two variants, and both need to be described in the procedure:

  • a single technician leaving — the vendor reports it within the deadline set in the contract, and the institution blocks that specific identity; the vendor’s other staff keep working,
  • the contract ending — blocking the vendor’s entire organisation, changing every credential its people may have learned, reviewing technical accounts created during the engagement and retaining the identity record and recordings.

For ICT services supporting critical or important functions, Article 28(8) of DORA requires exit strategies and transition plans. An exit plan usually describes the migration of data and services but skips the question of how to cut off the vendor’s people without interrupting the system. It is worth adding a list of accounts, keys and channels to close, with an owner for each item.

What does vendor access control look like with Fudo ShareAccess and Fudo Enterprise?

Fudo ShareAccess is a browser-based platform for sharing resources with third parties, designed so that the vendor needs no VPN, no agents and no infrastructure changes. It works together with Fudo Enterprise — a PAM platform that acts as the gateway to the institution’s resources. This is an example implementation of the pattern from the previous sections; the facts come from the vendor’s public documentation.

Connection direction. ShareAccess does not reach into the institution’s network directly. The connection is initiated by Fudo Enterprise, which builds a reverse SSH tunnel and needs outbound connections only — to the API over HTTPS on port 443 and, for the tunnel itself, on port 65522. The vendor notes that ShareAccess reduces reliance on a VPN in certain scenarios but does not take over every VPN function.

Identity and trust. Registration requires two-factor authentication (TOTP). An external user sees resources only once granted trusted status, and before granting it the administrator can compare the key fingerprint the user has provided. An invite option with immediate trust exists, but the vendor warns that it shares resources at once and without oversight — in a financial institution it is worth disabling it by policy.

Access for the task. Immediate access lasts a set time, up to 24 hours; scheduled access covers a defined from–to window. A justification for the request is mandatory, and the decision rests with the organisation owner. The user sees only the resources assigned to their safe (a group of resources).

Recording and revocation. External sessions can be monitored live and recorded in full. Recordings of SSH, RDP and VNC sessions stay in Fudo Enterprise inside the institution’s network; neither the ShareAccess platform nor the company operating it can see their content. Once a user is blocked in Fudo Enterprise, the shared resources become unavailable to them in ShareAccess as well.

In our assessment, this combination fits financial institutions with many vendors accessing critical systems well: a single place to revoke access and recordings kept inside the institution’s network answer the auditor’s two questions from the previous sections — who had access and whether it was withdrawn. Details are on the Fudo ShareAccess page.

How do you assess the maturity of vendor access control in a financial institution?

By checking which level four elements are at: identity, access time, recording and revocation.

LevelVendor identityAccess timeRecording and evidenceRevocationNext step
1. Contractualshared accounts, passwords known to the vendor teamstanding for the term of the contractsystem logs, if anymanual, when someone remembersinventory of accounts and channels
2. Organisednamed accounts in the most important systemsstanding, but reviewedlogin logs without session contentlist of accounts to close in the procedureaccess gateway for vendors with administrator rights
3. Controlledidentity in the gateway, target passwords in the vaulton request, with a time windowrecording with request metadatablocking in one placerights review every 6 months linked to the identity record
4. Demonstrableas above, plus verification when granting truston request, approved by the system owner”who, when, on what basis” report ready for the auditorbuilt into the contract exit plantesting the emergency path and the exit plan

The level is assessed separately for each vendor: the key integrator may be at level 3 while the hardware service firm is still at level 1. Priority is set by the criticality of the function that a given vendor’s access could disrupt.

Where do you start putting third-party vendor access in order?

With the list of vendors that hold privileged access to systems supporting critical or important functions — that list is shorter than the register of all ICT contracts, and the most depends on it. An order that structures the work:

  1. Inventory — vendor, system, accounts, connection channel, contractual basis and owner on the institution’s side.
  2. Closing side channels — standing VPNs, remote desktop tools and local accounts without an owner.
  3. Gateway and vault for the first group — vendors with administrator rights to critical systems move to access through a session broker, with target passwords in the vault.
  4. Request and approval — time windows, system owners as approvers, an emergency path.
  5. First rights review and a trial audit — a one-quarter report for one critical system, before the supervisor asks for it.

If you want to take this path with a team that combines PAM implementations with the requirements of the financial sector, see our PAM implementation and DORA readiness and compliance audit services. You will find more material in the DORA and identity and PAM hubs.

Frequently asked questions

Does DORA require a PAM system for vendors?

Regulation (EU) 2022/2554 does not name a tool. Delegated Regulation (EU) 2024/1774, however, requires in Article 21 that privileged access be assigned on a need-to-use or ad-hoc basis, that dedicated accounts be used for administrative tasks where possible, and — where feasible and appropriate — that automated solutions for privileged access management be deployed. With several vendors holding administrator rights, these requirements are hard to meet without such a tool.

Can vendor staff use a shared service account?

They should not. Article 20(2)(a) of Delegated Regulation 2024/1774 requires every staff member of an ICT third-party service provider who accesses the financial entity's assets to have a unique identity tied to a unique account. Article 21(c) requires generic and shared accounts to be limited as far as possible, so that every action can be traced to a specific person.

How often should vendor rights to critical systems be reviewed?

At least once every six months where a system supports critical or important functions, at least annually for other ICT systems, and whenever a change is needed — this is Article 21(e)(iv) of Delegated Regulation 2024/1774. Access that is no longer needed is withdrawn without undue delay, without waiting for the review.

Who is responsible for an incident caused from a vendor account?

Towards the supervisor — the financial entity. Article 28(1)(a) of DORA states that an entity using ICT services under a contract remains fully responsible at all times for meeting its obligations under the regulation. The contract may provide for claims against the vendor, but it does not transfer the regulatory obligations to them.

Does controlling vendor access require inbound connections and a VPN?

Not always. A vendor can connect through a session broker instead of a VPN into the internal network, and some solutions set up the tunnel from the institution's side as an outbound connection. The VPN does not disappear from every scenario, though — the decision is made separately for each type of access.

Share:

Talk to an expert

Have questions about this topic? Get in touch with our specialist.

Sales Representative
Łukasz Gil

Łukasz Gil

Sales Representative

Response within 24 hours
Expert consultation
Individual approach

Providing your phone number will speed up contact.

Want to Reduce IT Risk and Costs?

Book a consultation - we respond within 24h

Response in 24h Custom quote No obligations