One HTTP/2 Request Can Own Your Apache Server — CVE-2026-119618 Is the RCE Nobody Should Sit On
One HTTP/2 Request Can Own Your Apache Server — CVE-2026-119618 Is the RCE Nobody Should Sit On
The Apache HTTP Server project shipped version 2.4.71 on November 11, and if you run any release from 2.4.55 through 2.4.70 with HTTP/2 enabled, this is the update you should already have installed. The release fixes CVE-2026-119618, a critical flaw in the mod_http2 module whose description is unusually blunt: a race condition in the module's assign_headers() path lets an unauthenticated remote attacker corrupt heap memory and — in the worst case — execute code with the privileges of the web server process. No login, no session, no preparation: just the ability to reach the server over a connection speaking HTTP/2.
The flaw's lineage is unusual and worth noticing. It was reported on October 26 by Anthropic's OSS Scanner, the AI company's free scanning service that uses its models to audit open-source projects, with additional reporting from Anthony Weems and Simon Scannell of Google Cloud Vulnerability Research. A Linux vendor's assessment went further than most advisories dare: "public research has demonstrated a reliable exploit" against a default-configuration server with mod_http2 enabled. Wiz researchers confirmed the mechanics — improper synchronization when concurrent threads share a resource in the HTTP/2 stream handling lets an attacker trigger heap corruption that can be leveraged into RCE — while noting no in-the-wild exploitation had yet been observed. watchTowr, which reproduced the condition, put the guidance in four words: "patch now or drop h2 until you can."
Scope details matter for triage. HTTP/2 is only active where the Protocols directive enables it — many deployments that never turned it on are simply not exposed. Standalone mod_h2 builds (the GitHub-distributed module) are affected from 2.0.8 through 2.0.42, though the project repository has been archived and read-only since June, meaning downstream packagers are the patch path. Red Hat shipped fixes for RHEL 9 and 10 (RHSA-2026:98981 and RHSA-2026:98978) — but with an important nuance: Red Hat's own mod_http2 packages give each HTTP/2 stream its own memory allocator, which the vendor says makes exploitation much harder on RHEL than on upstream builds. Debian's tracker already carries the CVE. The 2.4.71 release also fixes a mod_md crash rated important and a low-severity one-byte stack overwrite in mod_vhost_alias.
What Should You Do?
- Inventory your HTTP/2 exposure before you panic. Check whether
Protocolsincludesh2on internet-facing virtual hosts — if the module isn't serving HTTP/2 anywhere, the RCE surface is closed and the update can follow your normal cadence. - Update to 2.4.71 on every server that does speak HTTP/2, prioritizing internet-facing hosts — unauthenticated, pre-handshake memory corruption on a public listener is a worm-class risk profile.
- Short-term mitigation if you can't patch today: remove
h2from the Protocols directive (forcing HTTP/1.1) — a performance trade, not an outage. - Don't trust "2.4.67 is recent enough." An earlier HTTP/2 flaw, CVE-2026-23918, was fixed in 2.4.67, and servers on that release are still in the affected range for the new CVE — check versions against the exact fixed list, not against "this year's updates."
The WAF Angle
This is a WAF-relevant RCE for one specific reason: the exploit rides the protocol layer your WAF may be terminating for you. When the WAF or reverse proxy in front of Apache speaks HTTP/2 to clients and clean HTTP/1.1 to the backend, the backend never sees the malformed concurrency that triggers the heap corruption — your architecture accidentally becomes the mitigation. But when Apache terminates HTTP/2 directly, a WAF in front of it inspects requests after protocol parsing has already happened, and a race condition inside the parser's own thread handling is not a payload any rule set can match. The practical takeaway: protocol-version control (which layer terminates h2) belongs in your hardening checklist next to rule tuning, and pre-auth memory-corruption bugs are the class where patch speed beats detection entirely. The same "the parser is the vulnerability" pattern drove Next.js's out-of-band RCE fixes and PHP2Shell's engine-level pre-auth RCE — when the bug lives below the application layer, the only defense that reliably works is not running the vulnerable code.