article://002-003/contractor-access-bastion

Building contractor access without handing over the MPLS

A dedicated entry point, named identities, temporary permissions and useful logs instead of broad network access.

This page is available in another language. FR

A contractor needs to maintain three devices. That does not mean they need to live on the whole MPLS. I want to replace the vague “connected to the network” with something exact: this person, these targets, this protocol, during this window. Nothing more.

The target path
  1. Identitynamed account and MFA
  2. Authorisationticket, scope and duration
  3. Bastioncontrolled crossing point
  4. Targetexplicitly allowed devices

Separate connection from authorisation

Signing in proves that an identity passed the expected control. It does not mean that identity should go everywhere. I split the decision in two:

  1. who is the person or team connecting?
  2. which targets and protocols are allowed for this intervention?

Shared accounts such as camera-contractor answer the first question badly. When five people use the same secret, the log tells us that an account connected, not who acted. Each operator needs a named identity or traceable federated identity.

MFA protects the entrance. It does not reduce network reach by itself. An authorisation policy still has to sit behind it.

A dedicated entry point instead of raw MPLS access

The target path crosses a separate access zone and a bastion, or an equivalent service. The external workstation gets no general route to the safety LANs. It reaches the entry point, then only the approved targets.

The bastion needs a concrete job:

  • centralise authentication
  • apply authorisation by group, target and protocol
  • restrict file transfer or clipboard use when the risk justifies it
  • record useful connection and activity data
  • remove access when the intervention window ends.

Dropping a VM called “bastion” into an open network achieves nothing. Routing and filtering must force the path through it.

Temporary permissions people can understand

A contractor permission should answer five questions without a novel:

Field Example decision
Who named identity or supplier group
What Web admin, SSH, RDP or required business protocol
Where exact device list or narrow segment
When time window and expiry date
Why ticket, incident or planned operation

The permission can be prepared in advance and switched on at the last moment. But it has to die for real at the end. A date in a ticket while an Active Directory group stays open for three years is not temporary access. It is permanent access with a pretty date next to it.

The external workstation stays external

Endpoint trust matters too. A perfectly authenticated account on a compromised device is still a problem.

Depending on the constraints, the design may require a managed workstation, check a few compliance signals or provide an isolated admin workspace. I prefer a realistic control we can test to a vague promise of a “secure laptop”.

Either way, the network should assume the endpoint may be hostile: no general discovery, no unnecessary routes, no direct path to other sites and only the protocols required for the job.

Log questions we may actually need to answer

I want to retrieve quickly:

  • who connected
  • which factor they used
  • from which context
  • which permission was activated
  • which target was reached
  • which protocol was used
  • when the session and permission ended.

Session video may help for some administration work, but it is not a reason to skip structured events. Searching two relevant minutes inside eight hours of video is miserable. Logs should narrow the search before anyone watches the details.

Retention must be proportionate, documented and compatible with internal rules. “Log everything forever” is not a strategy and may not be lawful.

Design emergency access before the emergency

Break-glass access may be necessary. It should be more visible, not more vague. The path needs a named authority, short duration, immediate alert and review after use.

If the only failure procedure is asking someone to open the whole MPLS temporarily, that temporary procedure will eventually become normal.

Acceptance tests

I would test at least:

  1. an authorised contractor reaching an authorised target
  2. the same contractor reaching a forbidden neighbouring target
  3. access after the time window expires
  4. a disabled account with an existing session
  5. loss of the bastion and the recovery path
  6. reconstructing the intervention from logs alone.

The test passes when the useful work gets through, everything else stops and the team can explain both. No fancier formula needed.

nav@bacchus:~

bacchus-dev.com navigation console · v1.0.0

Type help to list commands.

Tab autocomplete · ↑↓ history · Esc close · inputs stay local