Why You Should Switch Off Unused Legacy APIs
Unused legacy APIs add attack surface with no business value. Learn how to find them, why they stay enabled and what policy keeps them switched off.
Unused legacy APIs are old programmatic interfaces that stay enabled long after the integrations that needed them are gone. They rarely get attention, still run old code, and often expose a parser to the network. In November 2026 two of them, the Confluence Remote API and the N-able N-central Legacy API, became the entry point for the exploited Confluence CVE-2026-90436 flaw. Switching them off is one of the cheapest risk reductions available.
What is a legacy API?
A legacy API is an older interface a product keeps for backward compatibility after the vendor has moved to a newer one. Typical examples are SOAP or XML-RPC endpoints kept alive next to a modern REST API.
Vendors keep them because customers have scripts and integrations built on them. The vendor's development effort, testing and security review usually move to the new interface. The old one ages in place.
Why are unused legacy APIs a security risk?
They add attack surface that delivers no business value. If nothing uses an API, every flaw in it is pure risk with no offsetting benefit.
The Confluence case shows the pattern clearly. Atlassian turned the Remote API off by default for new installations in Confluence 3.0 and removed it entirely in 10.0. The setting still mattered in 2026 because instances upgraded over many years carried the old setting forward. The API ran on Apache Axis 1.x, a framework its own project no longer supports, and a flaw in it allowed unauthenticated code execution.
N-central followed the same shape. Rapid7 reports that the exploit chain against it depended on the non-default Legacy API being enabled, and N-able's November 16 hotfix disables that API.
Legacy APIs tend to share a few traits:
- Old dependencies. They are built on the libraries that were current when they were written, and those libraries reach end of life.
- Weaker authentication. Many predate token scopes, MFA-aware flows and fine-grained permissions.
- Less logging. Requests may not show up in the audit trail the way modern API calls do.
- No owner. Nobody on the current team remembers turning them on.
Why do they stay enabled?
Mostly because turning them off has a visible cost and leaving them on has an invisible one. Someone switched the API on years ago for a migration, a reporting script or an add-on. The person left, the script was retired, and the setting survived every upgrade. Disabling it means asking "will this break something?" and nobody has the answer, so the default is to leave it alone.
Defaults do not help here. Changing the default for new installs, as Atlassian did, protects new customers but does nothing for long-running instances.
How do you find legacy APIs in your environment?
Start with the products that hold the most sensitive data or sit closest to the internet: collaboration platforms, RMM tools, ticketing systems, identity products and file transfer servers.
- Read the admin settings. Look for options named "remote API," "legacy API," "XML-RPC," "SOAP," "classic API" or "compatibility mode."
- Check the vendor's deprecation notes. Release notes usually announce when an API is deprecated, disabled by default or scheduled for removal.
- Read the access logs. If an API endpoint has received no legitimate requests in 90 days, it is a strong candidate for removal.
- Map callers to owners. For every API that is in use, record which system calls it and who owns that system.
What should the policy be?
A short written rule works better than case-by-case debate. A reasonable starting point:
- Off unless justified. Legacy APIs stay disabled unless a named owner documents a current need.
- Expiry dates. Any exception gets a review date, tied to migrating the caller to the supported API.
- Upgrade checks. Each major upgrade includes a review of compatibility settings carried forward from older versions.
- Emergency authority. When a vendor publishes an exploited flaw in an optional feature, operations can disable it immediately and handle the broken integration afterwards.
That last point matters. For Confluence, Atlassian's own mitigation was to disable the feature, and our guide on how to disable the Confluence Remote API safely shows it takes minutes with no restart. Teams that already knew the API was unused could act the same day.
What does this cost compared with the alternative?
Disabling an unused API costs an hour of review and occasionally one broken script that should have been migrated anyway. An exploited remote code execution flaw on a server holding your internal documentation costs an incident response, credential rotation across every integration and possibly a breach notification.
There is also a quieter benefit. Every interface you remove is one less component to track when the next advisory lands. When a CVE names a library like Axis 1.x, a team with a short list of enabled interfaces can answer "are we exposed?" in minutes. For more on why flaws in shared libraries spread across products, see our coverage of the Atlassian CVE-2026-21589 shared library flaw. For the code-level weakness behind this case, read what unsafe reflection is and why it leads to code execution.