A jump server is a hardened intermediary that can provide a controlled, auditable path to private systems, reducing the need to expose RDP or SSH directly to the internet. The UK National Cyber Security Centre's case for building one is blunt: its Early Warning service alerted 67,000 IP addresses with internet-exposed RDP in 2022, and the NCSC identifies insecure RDP as a common ransomware initial-access route and one of the main remote-access tools exploited by attackers. This explainer defines jump servers against bastion hosts, privileged access workstations and cloud-native brokers such as Zero Trust architecture services, walks through a hardened SSH configuration, and sets out what UK Cyber Essentials actually expects before a jump host counts as compliant.
View the data behind this chart
| Who Operates It | Access Model | Example | |
|---|---|---|---|
| Jump Server | Admin team, self-managed | Shared host, manual hardening | In-house Linux VM |
| Bastion Host | Admin team or vendor | Single audited entry point | Hardened SSH/RDP gateway |
| Privileged Access Workstation | IT-issued to one user | Dedicated clean device | Locked-down admin laptop |
| Cloud-Native Identity Broker | Cloud provider, managed | No public IP, policy-based | Azure Bastion, AWS SSM, Teleport |
What a jump server actually is — and where "bastion host" fits
Strip away the branding and a jump server is an intermediary, usually hardened host or managed service, through which administrators reach private systems; organisations may operate multiple controlled paths rather than one universal machine. Logically, it sits in a perimeter network or DMZ, acting as the single inbound conduit between untrusted networks (such as the public internet or corporate LAN) and an isolated management VLAN containing sensitive servers. In a direct-access scenario, an administrator might expose port 3389 on a domain controller so they can work remotely, immediately exposing it to automated credential-stuffing attacks. Routing through a jump host closes that exposure: the domain controller accepts connections only from the jump server, requiring the administrator to authenticate at the hardened intermediary first with multi-factor authentication before jumping to the target machine. Microsoft's own marketplace listing for a bastion-host image describes this role: a single audited entry point into an Azure virtual network for access to private virtual machines.
The terms overlap: bastion host usually emphasises a hardened gateway, while jump server or jump host emphasises an intermediary hop. Usage varies by vendor and architecture. For instance, Microsoft positions Azure Bastion as a managed alternative to deploying and operating jump-host servers, providing RDP and SSH access to virtual machines without requiring public IP addresses on those VMs, removing the need for customers to maintain an underlying relay VM.
The distinction between traditional jump hosts and managed brokers matters for audits, though not because of the label. What an auditor actually evaluates is the retained evidence: who authenticated, with what factor, to which target, and what was logged. Either design may support audit objectives if it is appropriately configured and produces the evidence required by the relevant framework.

Why exposed RDP is the route ransomware actually uses
The NCSC's ransomware threat methodology is specific: attackers commonly use insecure RDP configurations — obtained through phishing, data breaches or credential harvesting — to get their first foothold, alongside exposed VPNs. This is not a theoretical risk category; it is the named mechanism behind the incidents the NCSC handles.
Scale matters here. The NCSC's sixth-year Active Cyber Defence report states that its Early Warning service sent alerts for 67,000 IP addresses with RDP exposed to the internet during 2022 — a count of IPs that triggered an alert, not a count of confirmed victims or active breaches. Separately, the NCSC's 2023 Annual Review names exposed RDP among the top five vulnerabilities notified through that same Early Warning service, confirming the problem persisted rather than being a one-year spike.
A different, earlier data point from a 2020 NCSC joint advisory records a 127% increase in exposed RDP endpoints — a measure of endpoint exposure during that period, not a ransomware-infection count, and on a different timeline from the 2022 figure above. The two should not be read as the same trend continuing; they are distinct measurements from different advisories.
Within the September 2022 to August 2023 window, the NCSC's Annual Review 2023 also recorded 297 reports of ransomware activity, of which 28 were triaged into NCSC-managed incidents. Again, these are two separate counts — reports received versus incidents the NCSC itself managed — and should not be conflated. The review also breaks reports down by sector.
Hardened setup: ProxyJump versus the agent-forwarding trap
A jump host is only as good as the SSH configuration behind it. The usual starting point on the jump server's own sshd_config is to disable root logins, disable password authentication entirely in favour of keys or certificates, restrict which accounts may connect at all, and cap authentication attempts — directives such as PermitRootLogin no, PasswordAuthentication no and AllowUsers combined with a short ClientAliveInterval to kill idle sessions.
On the administrator's own machine, the modern pattern is ProxyJump, configured once in ~/.ssh/config:
Host jumpbox
HostName jump.example.co.uk
User admin-jump
IdentityFile ~/.ssh/id_ed25519_jump
Host internal-db
HostName 10.0.4.12
User admin
ProxyJump jumpbox
IdentityFile ~/.ssh/id_ed25519_internalA connection then looks like a single command — ssh -J admin-jump@jump.example.co.uk admin@10.0.4.12. With ProxyJump, the jump host normally relays the connection while SSH authentication and encryption are established between the administrator’s client and the final host; disable agent forwarding unless it is explicitly required.
That is precisely what legacy agent forwarding (ssh -A) gets wrong. An attacker who controls the jump host while the agent socket is forwarded may be able to use that socket to authenticate as the administrator to reachable systems, subject to the agent’s restrictions, without ever touching the private key file itself. A single compromised jump host under agent forwarding becomes a master key to the entire estate it protects. ProxyJump normally avoids exposing the SSH agent socket to the jump host; keep private-key operations on the administrator’s device or approved hardware-backed credential system.
The NCSC reality: a jump box cannot inject trust it doesn't have
The most common mistake is treating a jump server as an all-encompassing security product rather than an access-control component. Microsoft's marketplace documentation for its bastion image is explicit on this point: a bastion design concentrates administrative access controls, authentication and audit records at a chokepoint, but it does not by itself eliminate the need to harden target systems or govern privileged accounts.
Put plainly — the jump host can verify who is connecting and log what they did, but it cannot retroactively make an unmanaged, unpatched laptop at the other end of that session trustworthy. If an administrator's own device is already compromised, routing their session through a pristine jump host doesn't remove the malware sitting on their keyboard. In practice, satisfying managed-device verification means validating endpoint posture before granting access — verifying MDM enrollment, disk encryption, an active EDR sensor, and hardware-bound certificates — rather than assuming any inbound SSH key belongs to a clean machine.
This is also the architectural seam where network microsegmentation and identity and access management start doing work a single jump host structurally cannot: verifying the device and the user on every request rather than once at the door.
Attack vectors: how a jump host becomes a lateral-movement hub
A jump host that is compromised stops being a chokepoint and becomes a launchpad, for three main reasons. First, agent-forwarding hijack, covered above — an attacker with control of the jump host's memory can ride the forwarded agent socket to every system the legitimate admin can reach. Second, in-memory credential harvesting: if a compromised jump host has access to cached credentials, session tokens or usable key material, those secrets may be abused; minimise and protect credentials stored or used on the host. Third, unpatched or exposed web management interfaces on the jump host or its surrounding appliance give attackers a route in that bypasses SSH hardening entirely.
Operational limitations must also be acknowledged: because a jump host concentrates administrative access controls and audit records at a single chokepoint, it inevitably becomes a high-value target for lateral movement. Breaching the jump host gives an attacker visibility into active management sessions and potential paths into every downstream target, turning a defensive gateway into a single point of catastrophic failure if host hardening and credential isolation fail.
The defence is to assume the host will eventually be targeted and ensure every session is logged and exportable. In the cited JumpServer implementation, documentation describes login and operation logs with these fields and export features; equivalent capabilities vary by product and configuration. Azure Bastion Premium can record graphical RDP and SSH sessions; verify current regional availability, storage configuration and associated charges before relying on the feature.
View the data behind this chart
| Reports received | NCSC-managed incidents | |
|---|---|---|
| Sept 2022-Aug 2023 | 297 count | 28 count |
The 2026 cost shape: self-hosted jump VM vs managed cloud bastion
There is no published UK pricing for jump server, PAM or ZTNA products in the sources behind this piece, so rather than guess at figures, it's worth being precise about the cost structure, which is what actually changes the decision.
A self-hosted jump VM's cost is mostly hidden in operations: someone has to patch the OS, rotate keys, review the session logs, and rebuild the host if it's ever suspected of compromise. The compute itself is typically the smallest line item. A managed service like Azure Bastion inverts that: Azure Bastion manages the Bastion service and connection relay, reducing the need to operate a customer-managed jump VM; customers still manage identity, permissions, target systems, logging choices and related configuration. However, Azure Bastion is a managed service with service-specific pricing; check Microsoft's current regional pricing, tier, deployment and data-processing charges before publication.
The honest procurement instruction is therefore: don't compare a VM's hosting cost against a managed service's headline rate. Compare the fully-loaded cost of the self-hosted option — including the labour of patching and log review — against the managed service's deployment and data-processing charges at your organisation's actual admin-session volume, and get a direct quote in GBP before committing either way.
UK compliance blueprint: Cyber Essentials Plus and audit evidence
The NCSC requirements describe examples of additional factors, such as a managed device, trusted-device app or separate token. Map the whole access design to the current Cyber Essentials requirements rather than treating any single jump-server pattern as sufficient. A jump server alone does not confer Cyber Essentials, Cyber Essentials Plus certification or legal compliance under UK GDPR; it is simply one access-control component within an in-scope estate.
Beyond the authentication factor itself, auditors and data protection governance both look for retained evidence rather than mere assertions. Audit logs and retained access-control evidence strengthen assurance and may support accountability obligations, but the exact evidence requirements should be checked against the applicable Cyber Essentials assessment and the organisation's GDPR governance.
Where supported, tools like JumpServer record login logs (success, failure, authentication method, source) and operation logs with field-level before-and-after changes that can be exported, while bastion host images on Microsoft Marketplace offer optional TOTP MFA for SSH logins and session auditing from first boot.
- •Map authentication factors to scheme guidance: managed device, trusted-device app or separate hardware token
- •Capture login logs: success/failure, authentication method, source address, failure reason
- •Capture operation logs: field-level before-and-after values for changes made during the session
- •Export logs to an external system so records survive any compromise of the jump host
- •Where target systems warrant post-event review, assess graphical session recording against tier and cost requirements
The decision: keep the hardened jump server, or move to PAM/ZTNA
None of this means the jump-server pattern is obsolete — it means it is the starting rung on a ladder. Modern designs increasingly combine bastion access with PAM, just-in-time permissions, private connectivity and ZTNA-style controls; suitability depends on the organisation’s architecture and risk.
A jump server may support an organisation’s broader access-control evidence, but it does not by itself satisfy or establish Cyber Essentials or Cyber Essentials Plus. It stops being adequate the moment the organisation has many administrators, many third-party contractors needing time-limited access, or a compliance requirement for session replay rather than connection metadata — at which point the conversation shifts toward ransomware protection built on managed detection & response and a proper PAM or ZTNA layer, not a bigger jump box.
Sources
Every figure in this article traces to the sources below.
- •Microsoft Marketplace — bastion host architecture and MFA/audit features
- •Microsoft — Azure Bastion product page and Premium session recording
- •NCSC — ransomware threat methodology and RDP/VPN attack vectors
- •NCSC — Active Cyber Defence sixth-year report, 67,000 exposed-RDP IP alerts
- •NCSC — Annual Review 2023, ransomware reports and top-five vulnerabilities
- •NCSC — joint advisory, 127% rise in exposed RDP endpoints
- •NCSC — Cyber Essentials IT Requirements v3.3, additional authentication factors
- •JumpServer documentation — login and operation logging detail
View the data behind this chart
| Layer | Detail |
|---|---|
| Academia | 50 reports, Sept 2022-Aug 2023 |
| Manufacturing | 28 reports, Sept 2022-Aug 2023 |
| IT | 22 reports, Sept 2022-Aug 2023 |
| Finance | 19 reports, Sept 2022-Aug 2023 |
| Engineering | 18 reports, Sept 2022-Aug 2023 |
