CVE-2026-129448: Apache Roller's XML-RPC Endpoint Hands Attackers Unauthenticated RCE
Apache is warning that attackers are already exploiting a critical flaw in Roller, its Java blog platform: an unauthenticated request to the XML-RPC endpoint deserializes attacker data before any login check, giving remote code execution. Here is how it works, who's affected, the fix, and how to tell if you were hit.
Executive summary. The Apache Software Foundation has confirmed that attackers are actively exploiting CVE-2026-129448, a critical (CVSS 9.8) unauthenticated remote code execution flaw in Apache Roller, the Java-based blog server. The XML-RPC endpoint (/roller-services/xmlrpc) deserializes "vendor extension" types before it authenticates the request, so an unauthenticated attacker can send a crafted XML-RPC request and run code on the server. Affected versions are commonly reported as 4.0.0 through 6.1.x; the fix is Roller 6.1.6 or later (some guidance points to 6.1.7). If you can't upgrade immediately, there is a config workaround, but note Apache's warning: turning XML-RPC off in the admin UI is not enough, because older code deserializes the request before it checks that setting. Patch now.
What it is
| Field | Detail |
|---|---|
| CVE | CVE-2026-129448 |
| Product | Apache Roller (Java blog server) |
| Type | Insecure deserialization (CWE-502) → unauthenticated RCE |
| Severity | 9.8 Critical |
| Vector | POST /roller-services/xmlrpc |
| Status | Exploited in the wild (Apache-confirmed); public reporting of active attacks |
| Affected | 4.0.0 – 6.1.x (ranges vary by source; confirm against Apache's advisory) |
| Fixed in | 6.1.6+ (6.1.7 current) |
How it works
Attacker ──► POST /roller-services/xmlrpc
│ XML-RPC body with a crafted
│ "vendor extension" type
▼
Roller deserializes the extension type
│ ← BEFORE authentication
▼
Gadget chain executes → code runs as the appThe root cause is a familiar one: deserializing untrusted input into rich objects. XML-RPC supports vendor-specific extension types, and Roller turned attacker-supplied bytes back into objects before confirming who was asking. That ordering, deserialize-then-authenticate, is what makes it unauthenticated.
Why it matters
- No credentials needed. Any host that can reach the XML-RPC path can be attacked.
- Internet-facing by nature. Roller runs public blogs, so the endpoint is often exposed.
- Code execution as the app. From there: web-shell, pivot, credential theft from config, defacement or watering-hole injection into the blog's own pages.
- It's being used now. This isn't theoretical; Apache says attacks are underway.
Remediation
Now
- Upgrade to Roller 6.1.6 or later (6.1.7 if available).
- If you can't patch immediately, apply the workaround and restart Roller:
- Set
enabledForExtensionstofalseinWEB-INF/web.xml. - Remove the
/roller-services/xmlrpcservlet mapping. - ⚠️ Disabling XML-RPC in the admin UI alone does not protect older versions, which deserialize before checking the flag.
- Set
- Restrict access to the XML-RPC path at the reverse proxy / WAF if the feature isn't needed externally.
Then
- Rotate secrets reachable from the Roller host (database credentials, API keys in config).
- Review the blog's rendered pages and templates for injected scripts or redirects.
Detection and hunting
Assume exploitation attempts; hunt from the disclosure window onward.
| Where | Look for |
|---|---|
| Web / proxy logs | POST to /roller-services/xmlrpc from unfamiliar IPs, odd/large bodies, bursts |
| Process telemetry | The Roller JVM (java) spawning sh, bash, curl, wget |
| Filesystem | New files in web-app directories (web shells), modified templates |
| Network | New outbound connections from the Roller host |
| Persistence | New cron jobs, systemd units, SSH keys, admin accounts |
A process-creation rule flagging the Roller JVM spawning a shell is the most reliable catch; see our blue-team thread on services spawning shells.
The pattern
Deserialization RCE keeps returning because the anti-pattern is easy to write and hard to spot: an endpoint turns attacker bytes into objects. We've seen it this month in Puppet Enterprise and the Apache Axis flaw in Confluence and N-central. The rule never changes: never deserialize untrusted input into rich objects, and authenticate before you parse.
Conclusion
CVE-2026-129448 is a textbook unauthenticated deserialization RCE in internet-facing software that is already being exploited. The fix exists, the workaround is specific (and the UI toggle is a trap on old versions), and detection is straightforward. Treat it as a same-day priority if you run Roller.
Do this now: upgrade Roller to 6.1.6+, or apply the web.xml workaround and restart; restrict the XML-RPC path; then hunt for the JVM spawning shells since the disclosure.
Related: CVE-2026-115731 Puppet Enterprise RCE and the network pentesting manual.
Sources: BleepingComputer, SecurityOnline, Security Affairs, William Ogou patch guide. Version ranges vary between sources; confirm against Apache's official advisory.
Related research
NGINX's HTTP/2 Reckoning: Two Critical Memory-Safety Flaws Open the Door to RCE
Five research teams independently found two critical memory-safety flaws in NGINX's HTTP/2 stack, one a use-after-free, that can lead to unauthenticated remote code execution on one of the internet's most-deployed web servers. Fixes shipped on 18 November. Here's who is actually at risk and what to do today.
"Am I Actually Patched?" The Apache mod_http2 RCE and the Distribution Patch Gap
A critical race-condition RCE in Apache's mod_http2 has a reliable exploit against default configurations, and it's paired with a nasty operational trap: your distribution may have fixed it without shipping version 2.4.71, so a version check can lie to you. Here's how the bug works and how to actually confirm you're patched.
n0n Names Visa and JPMorgan: A Case Study in Verifying Extortion Claims
A little-known extortion group listed Visa and JPMorgan on its leak site with a 48-hour countdown and no proof. Visa confirms an incident at a third-party provider but says its systems are untouched; JPMorgan blames a software bug. It's a perfect case study in how to read, and not overreact to, unverified breach claims.