Quarkus ships an emergency LTS release carrying 29 CVEs
Quarkus released 3.27.5.3 on 22 September 2026, described by the project as an emergency release for the 3.27 LTS stream. A matching emergency release went out for the 3.33 LTS stream.
The CVE list is the story. A few are in Quarkus itself:
- Quarkus Vert.x HTTP — authorization bypass through a path normalisation discrepancy in Quarkus HTTP security (CVE-2026-87743);
- Quarkus WebSockets Next — denial of service by out-of-memory through unbounded message buffering (CVE-2026-87742);
- Quarkus SmallRye Fault Tolerance — a memory leak in
@ApplyGuardleading to denial of service (CVE-2026-94449).
Most of the list is not Quarkus. Of the 29 entries, thirteen are Netty, and they cluster around HTTP protocol handling: request smuggling through incomplete validation of malformed Transfer-Encoding and through control characters in the chunk-size line; HTTP/1 authority-form CONNECT translated into a malformed HTTP/2 CONNECT with a Host-controlled :authority; HTTP/2 and HTTP/3 Extended CONNECT downgraded to regular CONNECT; unbounded queue growth in HttpServerCodec and in WebSocketServerExtensionHandler; an HpackEncoder denial of service on a large table size. Nine more are zstd-jni — use-after-free, out-of-bounds reads, a negative length parameter, a remote use-after-free in dictionary sharing.
The rest: Hibernate ORM SQL injection through an unescaped JSON path segment allowing data exfiltration and authorization bypass; Apache FreeMarker path traversal via a malformed locale identifier; WildFly Elytron password keyspace reduction through NFKC fullwidth folding; and SmallRye JWT unauthenticated same-origin SSRF through an unsanitised JWT kid header in AwsAlbKeyResolver.

What it means
Read the split before the total. Three of these are Quarkus bugs; the great majority arrived through the dependency tree, and no application team chose Netty or zstd-jni in the sense of evaluating them — they came with the framework.
That is exactly what an LTS stream is for, and it is also the argument for treating "emergency release" as a different class of event from a scheduled one. The HTTP smuggling cluster in particular is the kind that matters at a boundary: a proxy and an origin that disagree about where one request ends is a bypass of whatever sits between them, which is usually the authorization you trust.
The two Quarkus entries to read closely are the path-normalisation authorization bypass and the Fault Tolerance memory leak, because both are in code an application deliberately relies on for security and stability rather than in a transport it never thinks about.