announce@golang.org pre-announced Go 1.27.4 and Go 1.26.11 for US business hours on Tuesday, November 10, 2026. The releases include private security fixes covering CVE-2026-97033 through CVE-2026-97043 and CVE-2026-97054 (twelve CVEs). Binaries are not on go.dev/dl yet; stable remains 1.27.3 / 1.26.10. This closes the dated drop for the 10-27 heads-up that those CVEs were queued. The release dashboard is also stacking compiler wrong-code backports onto the same point releases (including the amd64/386 SBB carry-mask fix for #82825 and arm64 CSINV for #82777), so expect a fat security-plus-correctness drop.
π₯ What Changed
A quiet Friday with four items. Go pre-announced 1.27.4 and 1.26.11 for US business hours Tuesday 2026-11-10 with twelve private stdlib/toolchain CVEs (97033β97043 and 97054) β schedule the upgrade; go.dev/dl is still on 1.27.3/1.26.10. A new public gopls security issue, CVE-2026-97046, shows that require paths beginning with - in go.mod can be parsed as flags when gopls shells out to go list/go get/go mod why/go work use; no gopls release yet (still v0.24.0). Keycloak 26.8.2 is a dense Auth/IAM security train: account takeovers, unauthenticated SAML DoS/OOM, OIDC DCR/CIBA session issues, and the bundled Brotli OOB write. For Worth Your Time, Datadog's FETCh turns TLA+ specs into Rust runtime monitors on Husky routing and caught year-old conflicting placements that would have crashed backfill after placement data loss.
Also seen: gRPC C-core GHSA-v2cv-hgj4-927q (11-04 23:46 IST) β xDS routing used host while RBAC used :authority, so mismatched headers bypass host-based RBAC; High; patched 1.84.2/1.85.1; not grpc-go. Reducto r-1 mini ($1/1k pages) + r-1 GA and open-source ParseJudge (Apache-2.0); unset settings.model defaults to r-1 from 2026-12-07. AWS IAM Roles Anywhere now auto-fetches CRLs from certificate CDPs. oauth2-proxy v7.15.6 (Entra national-cloud tid, Go 1.26.10 bump).
Holds: Go 1.27.4/1.26.11 (pre-announced 11-10; still not on go.dev/dl); gopls release for #82821 CL 858103 + #82891 CVE-2026-97046 (proxy still v0.24.0); CL 858982 (#82853/#82861) still NEW; Go #82717 MPTCP default; CVE-2026-98432 Ubuntu still -proposed, OpenShift CoreOS still unfixed; x/oauth2 CL 854305 still NEW; Kafka 3.9.3 not released; moby/swarmkit GHSA IDs named in Docker 29 docs but advisory pages still 404; Postgres 19 RC2 ~11-12 / GA ~11-19. Alternates: IAM Roles Anywhere CDP (incremental), oauth2-proxy v7.15.6, Reducto (vendor OCR), #82888 html/template escaper (Go team: non-security).
PUBLIC-track Security issue. modfile.Parse permits require module paths that begin with a leading dash. When gopls passes those paths (or package/directory paths) as positional arguments to go list, go get, go mod why, or go work use, a leading-dash path is parsed as a command-line flag. Tracked as CVE-2026-97046. Milestone Unreleased (x/tools); latest gopls tag is still v0.24.0 (2026-10-15), which also still lacks the #82821 debug-bind fix (CL 858103 tip-only).
Keycloak 26.8.2 (2026-11-05) ships a dense security train across OIDC, SAML, Device Authorization, DCR, CIBA, UMA, FGAP, and Quarkus/Brotli. Highlights: CVE-2026-112208 account takeover via stale SSO_AUTH note in Organization identity-first login (reproduced on 26.8.1); CVE-2026-107123 / 107602 / 110931 unauthenticated SAML DoS, CPU exhaustion, and single-request OOM via SAML 1.1 parsers; CVE-2026-97174 CIBA sub injection issuing victim tokens; CVE-2026-97367 DCR claim substitution for Account REST session takeover; CVE-2026-107889 stored XSS on login via kcSanitize bypass; CVE-2026-117691 SAML redirect binding signature bypass via percent-encoded %53AMLResponse; CVE-2026-111497 bundled Brotli OOB write (same family as Node2Shell); plus device ACR bypass, offline-token revocation silently disabled on DCR RMW, theme gzip disk exhaustion, and session takeover via injected id_token_hint. Quarkus bumped to 3.40.1.1.
π Worth Your Time
FETCh (Formal Execution Trace Checker) compiles TLA+ specs into Rust runtime monitors that consume production telemetry and flag forbidden transitions. Applied to Husky (event store) routing with Router.tla (per-process sidecar; placements + hash) and PlacementStorage.tla (global non-overlap, RocksDB-backed). p90 sidecar overhead ~2% CPU / <0.2% memory across >1,300 pods at tens of millions of events/s. The placement monitor found intentional βskip dedup for balancingβ special cases that still persisted conflicting placements for about a year (~<0.01%); harmless in steady state, but two conflicts are enough to crash the backfill path after placement data loss. Specs refine into model-checkable forms; they proved consistent routing if both specs hold.