← Learn
Playbook9 min read

The end-of-life software exposure playbook — inventory, rank by exploitability, then decide ESU, isolate, or migrate

Published by Zero Hunt, an autonomous AI red team on an on-premise appliance running private AI: automated penetration testing for networks and infrastructure, black-box or gray-box, with a human approving every step that matters.

Short definition

A decision playbook for a standing population of unsupported, end-of-life software: how to inventory it, rank each asset by real exploitability and exposure, and choose ESU, isolation, or migration.

Why this matters now

End of life means no security patch will ever ship again — a standing attack surface that only grows. Several estates crossed fresh cliffs in October 2026 (Windows 11 24H2, the final paid ESU year for Windows Server 2012 R2) on top of Windows 10 ending support in October 2025 with roughly 40% of desktops still on it. Supervisors, insurers and NIS2 treat running unsupported software without compensating controls as negligence — so the question is not whether you are exposed, but which of these boxes is actually reachable and exploitable, and in what order you deal with them.

Key points

  • ▸EOL means no patch — ever. Once support ends the vulnerability backlog only grows; an unsupported box with no compensating control is a standing, permanent exposure.
  • ▸The cliffs are a rolling cadence: Windows 10 ended 14 Oct 2025, Windows 11 24H2 13 Oct 2026, and Server 2012 R2’s final paid ESU year 14 Oct 2026.
  • ▸Extended Security Updates buy time, not a cure. Consumer Windows 10 ESU runs to 12 Oct 2027; commercial ESU is $61 per device in year one and doubles every year for a maximum of three.
  • ▸Rank by exploitability, not asset count. A flat EOL inventory tells you what is unsupported, not what is reachable and exploitable — migrate by demonstrated risk, not alphabetically.
  • ▸Four moves per asset: migrate now, ESU-plus-migration, isolate-and-compensate, or accept with controls — chosen per asset by exposure × exploitability × business criticality.
  • ▸It is a compliance question too. NIS2 Art. 21, DORA, GDPR Art. 32 and cyber-insurance questionnaires all treat unmanaged EOL software as a failure of state-of-the-art controls.

Scope and when this playbook fires

Use this playbook for the standing, growing population of end-of-life (EOL) software in your estate — operating systems, server products, appliances, firmware, databases, and the libraries and components embedded inside applications for which the vendor has stopped shipping security updates. It fires whenever an asset crosses its end-of-support date, when an acquisition or a discovery scan surfaces unsupported systems you did not know you ran, or when a board or auditor asks “what are we doing about the Windows 10 boxes”.

The defining condition is that no patch is coming — by definition. That is what separates this from the KEV-driven emergency patch window: there a fix exists and the only question is how fast to deploy it; here there is no fix and never will be, so the decision is ESU, isolation, migration, or accepted risk under compensating controls.

Not in scope: routine lifecycle planning for software that is still supported (that is asset management, not exposure response); a supported product you have simply fallen behind on patching (use the patch-window playbook); or a confirmed intrusion through an EOL asset — at that point you are in incident response and the regulator notification clocks dominate.

The clock: EOL is a rolling cadence, and ESU only defers it

End of life is not a single event; it is a rolling cadence that never stops, and the last eighteen months concentrated several of the largest cliffs in computing:

  • Windows 10 reached end of support on 14 October 2025 — version 22H2 was its final release — yet it still ran on roughly 40% of Windows desktops at that date. Every Windows 10 machine not enrolled in Extended Security Updates has received no security fix since.
  • Windows 11 version 24H2 reached end of service on 13 October 2026, so fleets that moved off Windows 10 onto 24H2 must now take the next feature update or fall out of support themselves.
  • Windows Server 2012 / 2012 R2 left extended support in October 2023; its third and final paid ESU year ended on 14 October 2026, so as of this quarter those servers receive no updates by any route.
  • Exchange Server 2016 and 2019 reached end of support on 14 October 2025 — on-premises mail servers that are internet-facing by design.

Extended Security Updates (ESU) defer the cliff; they do not remove it. For Windows 10, the consumer ESU route extends critical and important security updates to 12 October 2027; the commercial ESU program runs for a maximum of three years, priced at $61 per device for year one and doubling every consecutive year — a deliberately escalating cost designed to push migration, not reward delay. ESU delivers security fixes only: no feature updates, no design changes, no support for anything else. Treat an enrollment as a funded, dated bridge to migration, not as a destination.

Phase A — build the EOL register and map exposure

Goal of the first window: know exactly what you run that is unsupported, and which of it is reachable. You cannot prioritize what you have not enumerated.

Checklist:

  • Build the EOL register. Enumerate every asset past — or within twelve months of — its end-of-support date: OS, server roles, appliances and firmware, databases, middleware, and the EOL libraries and components inside applications, where SBOM data earns its keep. Record vendor, product, version, end-of-support date, and ESU availability.
  • Attribute ownership. Every EOL asset gets a named owner accountable for its disposition. Orphaned assets are the ones that rot.
  • Map exposure. For each asset establish whether it is reachable from an untrusted network and what it connects to internally. Pull this from external attack-surface data and segmentation records, not from a network diagram that may be years stale.
  • Tag business criticality and dependency. Which business function does the asset support, and what breaks if it is isolated or migrated? An EOL box running a revenue-critical line-of-business app is a different decision from an idle print server.
  • Flag the industrial exception. OT, medical and embedded systems often run EOL operating systems by vendor design and cannot be patched or migrated on an IT timeline; they enter the compensating-control track from the start, not the migration track.

Phase B — rank by exploitability and decide the disposition

Goal: rank the register by real risk and assign each asset a disposition. The ranking input that separates a defensible program from a box-ticking one is demonstrated exploitability, not the raw count of unsupported machines.

Rank each asset on three axes:

  • Exposure — reachable from an untrusted network (highest), reachable only from a compromised internal foothold (medium), or well-isolated (lowest).
  • Exploitability — is there a known, weaponized path against the exact version you run, and is it actually reachable in your configuration? An internet-facing EOL mail server with a public exploit outranks an EOL internal appliance behind strict segmentation, even though both are “unsupported”.
  • Business criticality — the blast radius of both a compromise and the remediation.

The disposition decision tree, applied per asset:

  1. Migrate now — exposed and exploitable and high criticality. No ESU and no compensating control is an acceptable resting state here; this is the front of the queue.
  2. ESU + accelerated migration — you cannot migrate inside the window but the asset is exposed or important. Buy the dated ESU bridge and schedule the migration against its expiry, not against “someday”.
  3. Isolate and compensate — migration is blocked (an industrial system, an app with no supported successor) and ESU is unavailable or insufficient. The asset survives only behind compensating controls, with a documented residual-risk acceptance.
  4. Accept with controls — low exposure, low exploitability, low criticality. Record the decision and a review date; do not pretend it is fixed.

Write the disposition and a one-line rationale for every asset. That record is what an auditor, an insurer or a supervisor asks to see, and it is the line between “managed risk” and “negligence”.

Phase C — compensating controls and migration execution

For every asset not migrated immediately, a compensating-control set is the thing standing between an unpatchable box and a breach. Strongest first:

  • Network isolation and segmentation. Remove public reachability; place the asset in a tightly firewalled segment that can talk only to what it must. This is the single highest-value control for EOL assets and often the fastest to apply.
  • Strict egress control. An EOL box that cannot reach the internet outbound cannot easily be used to stage exfiltration or pull second-stage tooling.
  • Virtual patching. A WAF, IPS or protocol-proxy signature in front of the asset can blunt known exploit paths the vendor will never fix — a bridge, not a cure.
  • Access hardening. Force all administrative access through a jump host with MFA, strip standing credentials, and log every session.
  • Heightened monitoring. Watch the isolated segment and the asset’s behavior on the wire; an EOL system is exactly where you most want to catch anomalous activity in progress, because you cannot rely on it to defend itself.

Migration execution then runs against the ranked queue — highest risk first, each migration validated so that the successor actually carries the function and the old asset is fully decommissioned, not just powered down and forgotten (a dormant EOL VM is still an attack surface). Track each ESU enrollment against its hard expiry so the bridge does not quietly become the destination.

Evidence checklist — what to keep and in what order

Keep these artifacts ready, ordered by which gate consumes them — the goal is to prove to an auditor, insurer or supervisor that you run a risk-ranked, evidenced program, not an unmanaged pile of old software:

  • The EOL register — every unsupported asset with vendor, version, end-of-support date, owner, and ESU status.
  • The exposure map — which EOL assets are reachable, and from where.
  • The exploitability evidence — for the high-risk assets, proof of whether a known exploit path is actually reachable against your configuration, with dates.
  • The disposition record — migrate / ESU / isolate / accept, per asset, with rationale.
  • ESU enrollment records — which assets are covered, to what expiry, at what cost.
  • The compensating-control register — for every isolated asset, the controls applied and a review date.
  • The migration plan — ranked queue, target dates, decommission confirmation.

The recurring failure peer organizations report is not the migration work itself — it is defensibly ranking the register, because a flat list of unsupported assets says what is old, not what is dangerous. Establishing which EOL box is genuinely exploitable against its exact installed version is expensive to do by hand, and it is exactly what Zero Hunt’s AI Generative Pentest rail is built for: an autonomous AI red team writes a per-target exploit chain against the precise version an asset runs — black-box by default, gray-box against the installed source where you provide it — so “unsupported” resolves to “proven exploitable” or “proven not reachable”, and the migration queue is ranked by demonstrated risk rather than asset count. Because it runs on-premise on private AI, it can exercise a legacy estate without anything leaving the appliance — which matters precisely for the fragile, unpatchable systems you are least willing to expose — and its change-triggered campaigns keep the picture live as the estate shifts, with every finding signed at write time for the compliance evidence pack. It does not inventory your software or enroll your ESU; it answers the one question the flat register cannot: which of these actually gets you owned.

Common failure modes

1. Migrating by asset count instead of by risk. Teams clear the easy, low-risk EOL boxes to make the number go down and leave the genuinely exposed, genuinely exploitable system for last. Rank by exposure × exploitability × criticality, not by what is quickest to retire.

2. Treating ESU as a destination. An ESU enrollment is a funded, dated bridge. Buy it with a migration date scheduled against its expiry; an ESU that lapses with no successor in place is worse than no plan, because the budget was spent and the cliff still arrived.

3. “Powered down” is not “decommissioned”. A dormant EOL VM, an unplugged-but-still-joined appliance, or a forgotten DNS record pointing at a retired box is still attack surface. Confirm full decommission, not just that something was switched off.

4. Trusting a stale network diagram for exposure. The most common way an “internal, isolated” EOL asset turns out to be internet-reachable is an out-of-date diagram. Verify reachability with external attack-surface data; do not assume it.

5. Forgetting the EOL components inside supported apps. An application the vendor still supports can embed an end-of-life library or runtime. Without SBOM-level visibility these are invisible to an OS-level EOL scan, and they are a frequent breach path.

6. Never validating real exploitability. The most expensive failure at scale: pouring migration budget into EOL assets that are not actually reachable or exploitable while a genuinely owned one waits. Without continuous adversarial validation, the ranking runs on assumptions.

Cross-framework notes

Running unsupported software without compensating controls is treated as a failure of basic cyber hygiene across every regime a European CISO answers to:

  • NIS2 (essential and important entities) requires risk-proportionate technical measures under Article 21; an exposed, exploitable EOL asset with no compensating control is exactly the gap a competent-authority inspection looks for, and Italy’s d.lgs. 138/2024 carries it into national law.
  • DORA (financial entities) mandates ICT risk management for critical and important functions; an unsupported system underpinning such a function must be documented, compensated and on a migration path.
  • GDPR Article 32 requires “state of the art” technical measures; a regulator will read running known-exploitable EOL software that processes personal data as falling short of that standard.
  • Cyber insurance. Renewal questionnaires increasingly ask directly about end-of-life systems; running them without compensating controls can trigger a sub-limit, a higher premium, or a contested claim after an incident that traces back to an unsupported asset.
  • CISA and the FBI name the use of unsupported software as a product-security bad practice that is exceptionally dangerous for products supporting critical infrastructure — the clearest government statement that “it still works” is not a security position.

The discipline is the one every exposure playbook shares: decide once, evidence everywhere. The EOL register, the disposition record and the compensating-control register are the same artifacts your NIS2 risk-measures audit, your DORA resilience evidence and your insurance renewal all draw on — build the record once and export it into each frame.

Goes deeper

Want this against your environment?

Book a 30-minute scoping call — we will map this directly to your current compliance scope and threat profile.