CVE-2026-119618, the mod_http2 remote code execution bug that Apache fixed in httpd 2.4.71 on November 11, has moved from “patch soon” to “patch now.” Singapore’s Cyber Security Agency (CSA) issued an alert on November 16–17 saying the flaw is being exploited in the wild and that public proof-of-concept code is circulating. As of this writing, no second national agency has confirmed exploitation and the CVE is not in CISA’s KEV catalog. Even so, the bug is unauthenticated, reachable over the network, and present in every Apache build from the last two-plus years that serves HTTP/2. Don’t wait for KEV.
What the bug is
This is a thread-safety flaw in how mod_http2 handles concurrent HTTP/2 streams, not a classic use-after-free (CWE-362, improper synchronization). In httpd’s HTTP/2 architecture, a connection thread multiplexes the session, and stream requests are dispatched to worker threads. According to Red Hat’s analysis, a worker thread can end up writing into memory owned by the connection thread through a shared, lock-free allocator. The racy code path is in assign_headers(), which copies parsed request headers into the request’s header table. That code touches shared allocator state without proper locking.
When an attacker sends many concurrent streams on one connection, they can win the race and corrupt the heap of the httpd child process. Corruption that can be repeated and shaped this way leads to code execution as the httpd user, usually apache, www-data, or daemon. That account is often enough to read TLS private keys, if they’re readable by the worker, as well as proxied backend credentials, application config, and session material. It also gives a pivot point into the internal network behind a reverse proxy.
The flaw was found jointly by Anthropic’s OSS scanning effort and Google Cloud’s vulnerability research team. Apache rates it critical. Red Hat scores it 9.8. Singapore’s CSA uses CVSS 3.1 8.1, which reflects the attack complexity a race condition adds.
Affected versions
- Apache HTTP Server 2.4.55 through 2.4.70 with
mod_http2loaded andh2orh2coffered via theProtocolsdirective - Standalone mod_h2 2.0.8 through 2.0.42. The standalone repo is archived and gets no fix, so move to httpd 2.4.71’s bundled module.
- Servers that only speak HTTP/1.1 are not affected.
Distribution notes: Red Hat says the mod_http2 packages in RHEL 9 and RHEL 10 allocate each HTTP/2 stream from a separate allocator, which “substantially increases the complexity” of exploitation. Red Hat also says public research has shown a reliable exploit against a default-configured server with mod_http2 enabled, so treat RHEL as affected and patch it too. Check your distro tracker (Debian, Ubuntu, SUSE, Alpine) for backported fixes. A distro package version string won’t match upstream 2.4.71.
Exposure is broad. HTTP/2 is on by default in many vendor and container builds of httpd, including the official httpd Docker image configs, cPanel/Plesk stacks, and appliance web UIs built on Apache. It’s also common on reverse proxies in front of Tomcat, PHP-FPM, and internal APIs.
What to do now
- Upgrade to httpd 2.4.71 or your distro’s patched package, then restart. A graceful reload does replace children, but a full restart guarantees no old workers stay alive.
- If you can’t patch tonight, disable HTTP/2. Remove
h2andh2cfrom everyProtocolsline (global and per-vhost), or stop loadingmod_http2. LeaveProtocols http/1.1. Apache’s advisory doesn’t list an official workaround, but this removes the vulnerable code path entirely. If a CDN or load balancer terminates HTTP/2 and talks HTTP/1.1 to origin, origin exposure drops sharply. Confirm that’s actually true for your setup. - Find every instance. Run
httpd -vorapache2 -vacross your fleet, and check images and appliances that embed Apache. Verify HTTP/2 exposure externally withcurl -sI --http2 https://host/ -o /dev/null -w '%{http_version}\n', or by checking ALPN negotiation forh2. - Hunt. Look for repeated worker segfaults (
child pid ... exit signal Segmentation faultinerror_log) from a single source. Repeated crashes are the noisy side effect of losing the race. Also look for unexpected child processes spawned by httpd (sh,bash,curl,wget,python), new files in web roots or/tmpowned by the httpd user, and outbound connections from web-server hosts. Treat confirmed hits as host compromise: rotate TLS keys and any backend credentials the server can read.