
15 chapters · 13 min read
Just-in-time access: A CISO’s guide
What just-in-time access means, why most programmes stop at credentials and leave the network path standing, and the approvals, measures and evidence a CISO needs to prove access exists only when it should.
Written by Mark Fermor. Just-in-time access is one of the few security ideas that almost every framework agrees on and almost every organisation implements only halfway. This guide sets out what it means, where it is usually left incomplete, and how a CISO can prove that access which should not exist really does not exist.
Chapter 1Who this guide is for
This guide is for chief information security officers and heads of security who already run identity and privileged access controls, and who now need to reduce standing exposure across administrators, suppliers, engineering routes and automated systems. It assumes you know what privileged access management is. It concentrates on the decisions, the failure modes and the evidence.
Chapter 2The short version
- Just-in-time access means access is requested for a purpose, approved, granted for a limited period and then withdrawn. Nothing is left waiting in case it is needed.
- Most programmes apply it to credentials and privileges. Far fewer apply it to the network path itself, which means the route to a critical system often remains available long after the person using it has gone.
- That gap matters more each year. Attackers increasingly enter by exploiting software rather than by logging in, and an exploit does not wait for an approval workflow.
- The measure of success is not how many access requests you process. It is how many hours each critical system is reachable, and by whom, when nobody is working on it.
Chapter 3What just-in-time access actually means
The NCSC describes privileged access management as resting on two concepts: just in time administration and just enough administration. In the traditional model, possession of a credential is enough to reach a high-privilege interface. If that credential is stolen, the attacker inherits the same reach. Under just in time administration, the credential is instead used to request access. If the request is approved, a temporary credential is issued, and in the NCSC’s words it “will only work for a short period before it becomes useless”.
Just enough administration is the companion principle: when access is granted, it should carry only the permissions the task needs. The NCSC is direct about this: “Full, system wide access should be the exception, not the norm.”
The same idea runs through Zero Trust. NIST SP 800-207 states that access to individual enterprise resources is granted on a per-session basis. CISA’s Zero Trust Maturity Model describes the optimal identity stage as automation that authorises “just-in-time and just-enough access tailored to individual actions and individual resource needs”. Importantly, CISA applies the same language to the network pillar, describing optimal maturity as “dynamic just-in-time and just-enough connectivity for service-specific interconnections”. In other words, the frameworks already expect connectivity, not only credentials, to be granted just-in-time.
Chapter 4Why standing access is the problem
Standing access is access that exists whether or not anyone is using it: the administrator account with permanent rights, the supplier VPN that stays up between jobs, the remote support tool that was installed for one fault and never removed, the engineering route to a controller that nobody remembers opening. Each one is available to legitimate users and attackers alike, every hour of every day.
Verizon’s 2026 Data Breach Investigations Report, which analysed 2025 data, found that exploitation of vulnerabilities was the leading way into breaches at 31%, overtaking stolen credentials for the first time in the report’s 19 editions. SecurityWeek’s analysis of the report put credential abuse at 13%. Verizon also reported that breaches with third-party involvement rose by 60% and now account for 48% of all breaches.
Both findings point the same way for a CISO. Controls that only ask “who is logging in?” do not help when the attacker never logs in, and supplier connections are now involved in nearly half of breaches. The question that matters is simpler: is there a route to this system right now, and should there be?
Chapter 5The three layers of just-in-time access
It helps to treat just-in-time access as three separate layers. Each one answers a different question, and each one fails in a different way.
- Identity: who may ask. Strong, phishing-resistant authentication, named accounts and no shared logins. This establishes who is requesting access, but on its own it does not limit when or how much.
- Privilege: what they may do. Temporary elevation, scoped roles and credentials that expire. This is where most privileged access management products concentrate, and it is where the NCSC guidance sits.
- Path: whether a route exists at all. The network connection to the system is only available for the approved task and is withdrawn afterwards. If the route is not there, nothing can be sent across it, whether by an authorised engineer, a stolen session or an exploit against the gateway that would normally carry it.
Most organisations are mature at the first layer, partly mature at the second and largely absent at the third. Outside approved windows, the third layer is also the only one of the three that still holds when the remote access appliance, the jump host or the privileged access platform itself has a vulnerability. A temporary credential does not help if the attacker reaches the target through a flaw in the system that issues it.
Chapter 6Where just-in-time access matters most
You do not need to apply just-in-time access to every connection on day one. Start where a single compromise would give an adversary the most.
- Tier-zero administration. Directory services, identity providers, hypervisor and cloud control planes. The NCSC’s tiering guidance treats this as the root of trust that all other administration relies on.
- Backup and recovery administration. If the people who can delete your backups have standing access, so does anyone who takes over their accounts. Recovery is only credible if the recovery domain is not continuously reachable from production.
- Suppliers, integrators and managed service providers. The NCSC’s operational technology guidance classes connectivity from vendors, integrators and managed service providers as low trust, because you have limited control over their security. These routes are the clearest case for access that exists only for booked work.
- Operational technology and engineering access. PLCs, SCADA, building management, refrigeration and other controllers are often reached by vendors and engineers who need them occasionally but leave the route in place permanently. CISA’s OT fact sheet asks operators to configure remote access to the principle of least privilege “for the specific asset and user role or scope of work”, and to disable dormant accounts.
- Automated systems and AI agents. Service accounts, scripts and autonomous agents frequently hold the broadest standing access in the estate, because nobody wants a pipeline to fail on an expired permission. They deserve the same scrutiny as human administrators.
Chapter 7Designing the approval process
Just-in-time access is only as strong as the approval that releases it. The NCSC lists three broad approaches:
- Multi-party approval: access is released only when a different, authorised person approves it.
- Consensus: a group votes and access is released once a threshold is met.
- Rule-based automatic approval: access is released when defined criteria are met, for example a matching change ticket.
The NCSC is clear that the right level of approval is a risk management decision, based on the devices used and the impact if an attacker reached the interface. In practice, CISOs should set three or four approval tiers rather than one, and reserve two-person approval for the routes whose compromise would be most damaging. The process must also be fast enough that engineers do not work around it. An approval that takes a day will eventually be replaced by a connection that is never closed.
Decide, in advance, four things for every route in scope:
- Who may request it, and who may approve it.
- How long a window lasts by default, and who may extend it.
- What happens at expiry, including any safe-state requirement for operational systems that must not be interrupted mid-process.
- How emergency access works when the normal approvers are unavailable, and how that emergency route is itself kept closed until it is needed.
Chapter 8How just-in-time access fails in practice
Most just-in-time programmes do not fail at design. They decay. These are the patterns to look for:
- Approvals that are never refused. If every request is approved within seconds by the same person, the approval step is a formality rather than a control.
- Windows that never close. Extensions are granted repeatedly until a temporary route becomes permanent in all but name.
- Credentials expire, paths remain. The privileged session ends, but the VPN tunnel, firewall rule or remote support agent behind it stays in place, ready for the next session or for someone else.
- Standing service accounts. Human access is just-in-time, while the automation that does the same work holds permanent rights.
- Supplier tools outside the programme. Vendor-installed remote access software that bypasses your access platform entirely.
- Emergency access left open. Break-glass accounts and routes that are tested once, never rotated and permanently available.
- Closure that is assumed, not verified. A policy says the route closes at the end of the window, but nobody checks that it did.
Chapter 9What to measure
The number of access requests processed tells you that the workflow is busy. It does not tell you whether exposure has fallen. Measure these instead:
- Standing exposure hours: for each critical system, the hours per month it was reachable through each route, compared with the hours of approved work.
- Standing privileged accounts: human and non-human accounts that hold privilege permanently, and the trend over time.
- Supplier routes: the number of third-party connections that exist outside a booked window.
- Approval quality: the proportion of requests refused, queried or reduced in scope, and the median time to approve.
- Closure verification: the proportion of windows where withdrawal of the route was independently confirmed rather than assumed.
- Out-of-window activity: any attempted or successful access to an in-scope system outside an approved window, each one investigated.
Chapter 10The questions to ask your own team
- Which of our critical systems are reachable right now, and through which routes?
- Which supplier connections exist today with no work booked against them?
- When a privileged session ends, what else is withdrawn, and how do we know?
- Which service accounts and automated agents hold standing privilege, and why?
- If our privileged access platform or remote access gateway were exploited tonight, which systems could still be reached?
- How is emergency access kept closed until it is genuinely needed?
Chapter 11The evidence to expect
- An inventory of in-scope routes, each with an owner, approval tier, default window and expiry behaviour.
- Measured standing exposure hours for the most critical systems, reported monthly.
- Approval records showing who requested, who approved, the scope granted and when the access ended.
- Independent confirmation that routes were withdrawn at the end of each window.
- A tested emergency access procedure, with the date it was last exercised.
- A residual risk statement for any route that must remain permanently available, accepted by the accountable owner.
Chapter 12A practical sequence for the first 90 days
- Weeks 1 to 4: find the standing routes. Inventory privileged accounts, supplier connections, remote support tools, engineering routes to operational systems and automated identities. Expect to find more than the documentation shows.
- Weeks 5 to 8: tier and decide. Rank routes by the impact of compromise. For each one, decide whether it should be removed, made just-in-time or accepted as a documented exception.
- Weeks 9 to 12: pilot and measure. Apply just-in-time access to one high-impact group, typically supplier access or backup administration. Measure standing exposure hours before and after, and report the change.
Extend from there in order of impact. Each wave should reduce a number the board can see.
Chapter 13What just-in-time access does not do
Just-in-time access reduces the time an attacker has. It does not remove the risk inside an approved window, when the route and the privilege both exist. It does not protect you from an approver whose own account has been compromised, which is why the highest-impact routes justify two independent approvals. It does not replace patching, monitoring or incident response. And it does not, on its own, inspect or validate what crosses a connection while it is open. State these limits plainly in your residual risk statement; they are where the next control has to come from.
Chapter 14Where Firevault fits
Identity and privileged access platforms handle the first two layers well, and Firevault does not replace them. Control by Firevault addresses the third: it governs whether selected physical network connections to critical systems exist at all. Control is delivered on premises, and it applies just-in-time access at the connection itself rather than relying only on a software rule.
- Lock allows selected connections only during approved time windows, with defined expiry and closing behaviour, subject to safe-state requirements.
- Unlink physically withdraws selected standing access paths, so critical systems are not continuously reachable through them until a separate approved need justifies restoring the route.
- Validate requires two independent approvals for the same proposed connection change before it is released.
- Relay stages approved connectivity between public, intermediate and private zones, so third-party and remote access does not leave an open end-to-end route.
These Modules come together in Control Plans built around the situations in this guide: Privileged Infrastructure Access, Third-Party Access Control, Scheduled Critical Maintenance and Industrial Engineering Access.
The same principle applies to data. Offline Secure Storage® keeps a defined set of critical data physically disconnected by default, connected only for authorised access, so recovery data is not continuously reachable from the estate it is meant to restore.
If you want to see which of your routes could become just-in-time first, start with Control for CISOs.
Chapter 15Sources
- NCSC, Secure system administration: Use privileged access management. Just in time and just enough administration, and approval processes.
- NCSC, Secure system administration: Risk manage administration using tiers. Tier 0 as the root of trust.
- NCSC, Operational Technology: Understand and document third-party risks to your OT system. Trust levels for vendor, integrator and managed service provider connectivity.
- CISA, Zero Trust Maturity Model, Version 2.0 (April 2023). Just-in-time and just-enough access in the identity, network and data pillars.
- NIST SP 800-207, Zero Trust Architecture. Per-session access to enterprise resources.
- CISA, FBI, EPA and DOE, Primary Mitigations to Reduce Cyber Threats to Operational Technology (May 2025). Least-privilege remote access and dormant accounts.
- Verizon, 2026 Data Breach Investigations Report findings (19 May 2026). Vulnerability exploitation at 31% and third-party involvement at 48%.
- SecurityWeek, Verizon DBIR 2026 (20 May 2026). Credential abuse at 13%.
Where Firevault fits
Controls an auditor can physically verify
Firevault gives you physical separation, named custody and evidenced access, so compliance claims about isolation and control are things you can show, not just assert.
- 1Custody Named, access-controlled hardware in a Firevault Bunker
- 2Evidence Access windows and retrieval events are recorded
- 3Separation Physical isolation that satisfies offline copy requirements
- 4Jurisdiction Stored where your regulatory position requires



Put this guide into practice
Ready to apply what you have learned? Explore how Control by Firevault governs the physical paths into your systems.
From £360 a month including VAT. 36-month commitment. First payment at checkout.

Business
Size your Vault.
2TB, 4TB or 8TB of dedicated hardware, from £360 a month inc VAT.

Larger estates
Request a Storage design.
From 20TB, designed with our solutions team.

Personal
LUV, 300GB.
Identity documents, wills and family archives, opened one day a week.



