Choosing between an SSH bastion host and a VPN for server access is not simply a question of which tool is more secure. The right design depends on how narrowly you can define access, how users and devices are verified, what needs to be logged, and how the environment must recover when an identity provider, network route, or access service fails. This guide compares both architectures, explains when using them together makes sense, and provides a recurring workflow for reviewing secure SSH remote access.
Overview
An SSH bastion host, sometimes called a jump host, is a controlled intermediary through which administrators connect to private servers. Instead of exposing every server to the internet, an organisation exposes or otherwise makes reachable a hardened access point. The user authenticates to the bastion, then connects onward to an approved destination.
A VPN creates an encrypted network path between an approved device and a private network, cloud environment, or segment of an organisation's infrastructure. Once connected, the user may be able to reach multiple internal services, depending on routing and firewall rules. A VPN therefore operates primarily at the network-access layer, while a bastion host can provide a more narrowly controlled path to specific administrative sessions.
Neither design is automatically least privilege. A VPN that places every remote user on a broad internal network can create unnecessary reach. A bastion host with shared accounts, unmanaged keys, or unrestricted forwarding can become a high-value target. The security outcome depends on identity controls, device posture, segmentation, logging, patching, and the quality of access reviews.
For a small environment, a bastion may be the simpler way to centralise SSH access. A VPN can be useful when staff need several internal protocols, private service discovery, or access to systems that are not administered over SSH. Many mature designs use both: the VPN restricts network reachability, while the bastion concentrates privileged SSH access and session controls.
What to track
Exposure and network reach
Record which systems are reachable from the public internet, from the VPN address space, and from the bastion. The preferred exposure is not always identical for every workload, but public SSH access should be treated as an explicit exception rather than a default. Review firewall rules, security groups, routes, and forwarding settings together; a service that is not directly exposed may still be broadly reachable through an over-permissive VPN or jump host.
Track whether users receive access to a whole subnet or only to named hosts and ports. If a VPN is retained for general internal access, consider separate groups, network segments, and destination rules for administrators, developers, contractors, and automated workloads.
Identity and device trust
Compare how each design integrates with the organisation's identity provider. Look for individual accounts, multi-factor authentication, short-lived credentials, group-based authorisation, and prompt offboarding. SSH keys should belong to a person or workload identity, not to a team, and should have an owner, purpose, creation date, and rotation or expiry plan.
Device trust is a separate question from user identity. A valid account does not prove that the connecting laptop is patched, encrypted, managed, or free from suspicious tooling. Track whether the VPN or bastion workflow can enforce device certificates, endpoint-management status, operating-system requirements, or a separate privileged access workstation policy. Where such signals are unavailable, document that limitation rather than treating MFA as a complete control.
Session visibility and recording
List the logs produced by the VPN, bastion, operating system, cloud platform, and identity provider. At a minimum, you should be able to associate a connection with an individual or workload, a source device, a destination, and a time window. Command logging or session recording can improve investigation and accountability, but it must be implemented carefully: sensitive command arguments, secrets, and personal data may appear in logs.
Check retention, access permissions, time synchronisation, alerting, and whether logs remain available if the bastion or VPN service is compromised. A record that exists only on the accessed server may not be sufficient for incident investigation.
Automation and operational access
Human administrator access and automated deployment access should not share credentials. Track CI/CD runners, configuration-management systems, backup jobs, and monitoring agents separately. Prefer narrowly scoped identities, restricted source locations, non-interactive keys or certificates where appropriate, and explicit destination allowlists. Do not route automation through a user VPN simply because it is convenient.
For human access, test normal SSH workflows, bastion proxy commands, certificate renewal, and emergency access. For automated access, test key expiry, runner replacement, host-key changes, and failure of the identity or secrets-management dependency.
Controls and recovery
Maintain a bastion host security checklist covering operating-system updates, minimal installed services, administrative ports, outbound connections, host-key management, malware monitoring, backups, and recovery images. For VPN access, track client versions, cryptographic configuration, split-tunnel rules, DNS handling, kill-switch behaviour where relevant, and account-session controls.
Review the recovery path before an outage occurs. Can administrators reach the environment if the VPN control plane is unavailable? Is there a break-glass account protected by an independent method? Can the bastion be rebuilt from a known configuration rather than repaired indefinitely? Emergency access should be tightly monitored, time-limited, and reviewed after use. For account-related incidents, use a documented response process such as the compromised VPN account response plan.
Cadence and checkpoints
A useful review is recurring rather than triggered only by a suspected breach. Use a monthly operational check for high-change environments and a quarterly architecture review for most teams. Record the date, reviewer, systems covered, exceptions, and actions assigned.
Monthly checks
- Review new, removed, and dormant VPN accounts, SSH keys, certificates, and bastion users.
- Compare firewall and security-group changes with approved tickets or infrastructure code.
- Check failed logins, unusual source locations, unexpected destinations, and use of emergency access.
- Confirm that bastion and VPN software, operating systems, agents, and clients are within the organisation's supported patch policy.
- Test a normal administrator connection and confirm that required logs arrive in the intended monitoring system.
Quarterly checkpoints
- Rebuild the access matrix from current roles rather than copying the previous quarter's list.
- Ask each system owner whether every server, port, group, and automation identity is still required.
- Test offboarding, credential revocation, certificate expiry, and recovery procedures.
- Review whether a broad VPN route can be replaced with a narrower bastion, application, or zero-trust access rule.
- Sample session records and verify that timestamps, identities, destinations, and retention meet the team's investigation needs.
When the architecture is changing, increase the frequency temporarily. A cloud migration, new identity provider, network redesign, merger, or major SSH or VPN upgrade can invalidate assumptions that were reasonable at the previous review. The broader context is covered in the secure remote access checklist and the comparison of ZTNA versus VPN for remote teams.
How to interpret changes
When access requests increase, do not respond by automatically expanding the VPN. First identify what users actually need: a shell on one host, access to a database, an internal web application, or several services across a private network. A bastion may suit narrowly defined SSH administration, while a VPN may be justified for multiple protocols or legacy systems. If users need one application rather than network access, an application-level or zero-trust design may reduce reach further.
If session logs are incomplete, treat that as an architecture problem rather than merely a monitoring problem. A design that cannot reliably identify the person behind a privileged connection makes investigation and access review harder. Improve identity integration, certificate issuance, central logging, or session controls before adding more users.
If emergency access is used repeatedly, investigate the underlying failure. Frequent bypasses may indicate unreliable VPN capacity, an unavailable identity provider, an impractical MFA workflow, or a bastion that is too restrictive for legitimate operations. The answer should be a controlled design improvement, not permanent broad access.
Scoring can make reviews consistent. For each access path, rate exposure, identity assurance, device assurance, least privilege, logging, automation separation, operational overhead, and recovery on a simple internal scale. Keep the reasoning beside each score. A lower score is not automatically unacceptable if the path is isolated, monitored, temporary, and supported by compensating controls; an apparently strong score is not meaningful if the evidence is out of date.
When to revisit
Revisit the SSH bastion host versus VPN decision at least quarterly, and sooner after a security incident, privileged-account review, cloud redesign, major platform release, or change in identity provider. Also review it when a team adds contractors, introduces new deployment automation, adopts a different endpoint-management model, or discovers that users are sharing keys or bypassing the approved route.
At the next checkpoint, begin with a current asset and identity list. Draw the real connection paths, including temporary rules and emergency routes. Remove unused keys and accounts, narrow destinations, confirm MFA and device requirements, test logging, and run one recovery exercise. Capture unresolved exceptions with an owner and expiry date.
The practical goal is not to choose a fashionable architecture. It is to make every privileged connection deliberate, attributable, limited to what is needed, and recoverable when a dependency fails. For many teams, a hardened bastion combined with carefully segmented VPN access is a reasonable starting point. As identity, device trust, and application controls mature, the same quarterly review can show where that design should be narrowed, replaced, or extended.