A Brotli decoder out-of-bounds write that reaches far past Node.js. The vulnerability is real and patched; the RCE claim is contested and reproduced twice; in-the-wild exploitation is now confirmed as attempts (Zscaler 402 decoy events; Datadog confirmation), with no successful production RCE and no victim named to 10 November. Three claims, three different confidences — the 28 October update moves one of them.
Out-of-bounds write in Google's reference Brotli decoder, c/dec/decode.c, present since 2018 and fixed upstream in February 2026.
CVSS 4.0 9.2 from the CNA; Node.js rates it High; NVD ingested 22 Oct, Analyzed by 28 Oct, 9.8 / 9.2.
Two teams report RCE on stock builds. Imperva on Linux arm64, ELLIO on x86-64 across four frameworks.
Both describe it as probabilistic, offset-specific and crash-heavy. At least one researcher argues the write is unreachable and the real impact is denial of service.
One observed exploit attempt (26 Oct), which failed. No victim, no campaign, no actor.
In CISA KEV since 27 Oct (remediation due 10 Nov) on evidence of active exploitation; CISA SSVC moved from none to active. The attempt failing does not make the listing wrong — it makes it a lower bound.
A crash in a Node.js worker is easy, reliable and public. Code execution is neither. Everything in the PoC ecosystem that "works" mostly works by killing the process — and because failed exploitation attempts also crash workers, an attacker who cannot get a shell still has a denial-of-service primitive. Patch on the RCE claim's plausibility, not on its proof.
Splunk’s advisory SVD-2026-1015 was updated on 17 November — its first movement since 9 November. Splunk Enterprise 10.0 includes Node.js with the affected Brotli package, used only by the legacy Simple XML PDF renderer, which is disabled by default and unused; versions 9.4 and 10.2 and higher do not include the vulnerable package, and Splunk recommends upgrading to 10.2 or higher. The Splunk Observability OpenTelemetry Collector fix is now named: upgrade to 0.165.0 or higher. Cisco’s October Node.js advisory moved to v1.14 on 17 November, with the same change note as v1.13; the live advisory confirms IOS XE, IOS XR, ASA and FMC as Not impacted, as already recorded under v1.12 — which revision first moved them still cannot be pinned from Cisco’s history. No new Node.js release in the window, and no new exploitation evidence for the November batch.
Splunk Enterprise 10.0 bundles the vulnerable Brotli package only for the legacy Simple XML PDF renderer, disabled by default and unused; verify the legacy-export feature flag and plan the upgrade to 10.2 or higher, which does not include the vulnerable package. Splunk plans to upgrade the packages in a future 10.0 release, so 10.0 stays affected until then. The Visual Exporter app still bundles the affected package in all Enterprise versions — disabling or removing it breaks scheduled dashboard exports.
Sources: Splunk — SVD-2026-1015 (Last Updated 17 Nov 2026) · Cisco — Vulnerabilities in Node.js and Brotli Affecting Cisco Products: October 2026 (v1.14, 17 Nov 2026)
| Evidence | Status to 18 Nov |
|---|---|
| Splunk advisory SVD-2026-1015 (updated 17 Nov) | Updated 17 Nov — first movement since 9 Nov. Splunk Enterprise 10.0 includes Node.js with the affected Brotli package (legacy Simple XML PDF renderer only, disabled by default, unused) — new detail; 9.4 and 10.2+ do not include the vulnerable package — new detail; upgrade to 10.2 or higher remains the recommended path. Splunk plans to upgrade the packages in a future 10.0 release. Observability OpenTelemetry Collector: upgrade to 0.165.0 or higher (new). Synthetic Runner 1.71.0 or higher (unchanged); Galileo remedied. Enterprise, Visual Exporter, SOAR Automation Broker, UBA, Attack Analyzer, Python for Scientific Computing, AppDynamics SaaS/On-Premises, Smartlook and Observability Cloud remain In progress. |
| Cisco advisory v1.14 (17 Nov) | Version bump; the change note is identical to v1.13 (“Updated products under investigation, vulnerable products, and products confirmed not vulnerable”). The live advisory confirms IOS XE, IOS XR, ASA and FMC Not impacted — as already recorded under v1.12. Which revision first moved them cannot be pinned from Cisco’s history, since v1.13 and v1.14 carry the same note. Status still Interim; PSIRT not aware of malicious use; PoC code available. Still covers the October CVEs only — no November-16-CVE advisory yet. |
| Node.js November releases | No new release in the window — still v26.13.1, v24.22.3, v22.23.5 (11 Nov). |
| CVE-2026-123873 | Still no NVD record (checked 18 Nov). CVE-2026-106071 is confirmed a distinct October item (NVD: published 23 Oct, last modified 2 Nov, 26.x only, 3.6 Low, wildcard --allow-fs-read / --allow-fs-write over-grant) — it does not close the 123873 gap, which stays open. |
| Neat advisory NEAT-SA-2026-11-02 | Unchanged — still “Interim: fix in progress” (2 Nov); Neat OS 26.6.3 Stable rollout completes by 21 Nov. |
| Confirmed production compromise | None named to 18 November. This row is empty on purpose — do not add attempts + KEV into a victim count. |
Sources: Splunk — SVD-2026-1015, Last Updated 17 Nov 2026 (Enterprise 10.0 impact; Otel Collector 0.165.0) · Cisco — Vulnerabilities in Node.js and Brotli Affecting Cisco Products: October 2026, v1.14, 17 Nov 2026 · NVD — CVE-2026-106071 (3.6 Low; published 23 Oct; 26.x only)
Neat OS 26.6.3, the release carrying the Brotli fix for CVE-2026-111497, reached the Stable channel on 7 November; the staged rollout completes by 21 November. Neat’s own advisory NEAT-SA-2026-11-02, published and last updated 2 November, still reads “Interim: fix in progress” and describes the fix as “planned for OTA Stable release on November 21, 2026” — the advisory has not been updated since the release shipped, so its “planned” wording is stale against Neat’s own software-updates page, which lists 26.6.3 as the latest Stable release since 7 November. Neat reports no evidence of exploitation against Neat devices. UC Today reported the rollout on 13 November. No new Node.js release in the window, and no new exploitation evidence for the November batch.
NEAT-SA-2026-11-02 still says the fix is “planned” for 21 November, because it has not been revised since 2 November. Neat’s software-updates page shows 26.6.3 as the latest Stable release, published 7 November, with the staged rollout completing by 21 November. Treat 26.6.3 as available now — staged, so a device with automatic updates off, or not yet reached in the rollout, is still unpatched.
Sources: Neat — Security advisory November 2026, Brotli decoder vulnerability CVE-2026-111497 affecting Neat devices (NEAT-SA-2026-11-02) · Neat — Software updates (Stable latest 26.6.3, 7 Nov 2026; Preview 26.6.4, 14 Nov) · UC Today — Weekly Wrap, 13 Nov 2026 (Neat rolls out fix for Brotli flaw)
| Evidence | Status to 17 Nov |
|---|---|
| Neat OS 26.6.3 (Stable) | Shipping since 7 Nov 2026; staged rollout completes by 21 Nov. Carries the Brotli fix for CVE-2026-111497. Advisory NEAT-SA-2026-11-02 (2 Nov) still reads “Interim: fix in progress” and “planned for OTA Stable release on November 21, 2026” — not updated since the release shipped. Neat reports no evidence of exploitation against Neat devices. |
| Neat OS 26.6.4 (Preview) | Preview channel, 14 Nov 2026. The Stable line remains 26.6.3. |
| Node.js November releases | No new release in the window — still v26.13.1, v24.22.3, v22.23.5 (11 Nov). The 16-CVE batch stands as recorded in the 12–13 November sections. |
| CVE-2026-123873 | Still no NVD record (checked 17 Nov). The 16-versus-17 count gap and the CVE-2026-106071 description match from the 13 November section stay open. |
| Cisco advisory v1.13 (13 Nov) | Version bump only. The change note is the same boilerplate as v1.12 (“Updated products under investigation, vulnerable products, and products confirmed not vulnerable”); the substance recorded under v1.12 stands, including ISE Not impacted. Still covers the October CVEs only — no November-16-CVE advisory yet. |
| Splunk advisory SVD-2026-1015 | No new revision in the window — last update 9 Nov, documentation-only. |
| Confirmed production compromise | None named to 17 November. This row is empty on purpose — do not add attempts + KEV into a victim count. |
Sources: Neat — Security advisory November 2026, Brotli decoder vulnerability CVE-2026-111497 affecting Neat devices (NEAT-SA-2026-11-02) · Neat — Software updates · UC Today — Weekly Wrap, 13 Nov 2026
The CVE records for the 11 November batch now carry CNA scores for eight of the nine remaining CVEs: all six Medium CVEs score 4.2–6.8 and two of the three Low CVEs score 2.5 and 2.9, published to NVD on 11 November and last modified 12 November. The highest is CVE-2026-123877 at 6.8 (https.Agent TLS session-cache policy bypass); the lowest scored is CVE-2026-123829 at 2.5. The scores are the OpenJS Foundation CNA’s, not NVD’s — every record still reads “Awaiting Analysis” with no NVD assessment. This supersedes the 12 November evening section’s “no score verified” for the Mediums and two of the Lows; CVE-2026-123873 remains unscored. Cisco’s Node.js advisory moved to v1.12 on 12 November (products under investigation, vulnerable products and products confirmed not vulnerable updated; still the October CVEs only). Apple released HomeView Software 27.2 on 12 November, and its security content lists the Brotli CVE-2026-111497 fix — the first HomeView release to carry it. No exploitation statement, PoC or KEV listing was found for the November batch in this check.
CVE-2026-123873 (Low, Permission Model heap-snapshot destination check, listed in all three release notes) still has no NVD record (checked 13 November), so the advisory’s sixteen descriptions against the release notes’ seventeen CVE IDs remain unresolved. Separately, the advisory’s wildcard --allow-fs-read over-grant description matches the NVD text of CVE-2026-106071 (3.6 Low) word for word — but that record was published 23 October, lists Node.js 26.x only and references the October releases, while the advisory says the wildcard flaw affects all lines (22.x, 24.x, 26.x). Whether the November advisory item is that CVE or a separate November assignment is not established by the sources read; do not merge them, and do not treat the 106071 record as closing the 123873 gap.
Sources: NVD — CVE-2026-123877 (6.8) · NVD — CVE-2026-123871 (6.2) · Node.js, Wednesday, November 11, 2026 Security Releases
| Evidence | Status to 13 Nov |
|---|---|
| The six Medium (CVSS, CNA) | CVE-2026-123877 6.8 (CVSS 3.0; https.Agent TLS session-cache policy bypass); CVE-2026-123871 6.2 (CVSS 3.0; proxy credentials in ERR_PROXY_INVALID_CONFIG); CVE-2026-123828 5.9, CVE-2026-123868 5.9, CVE-2026-123875 5.9 (CVSS 3.0; HTTP/2 Alt-Svc one-byte overflow; HTTP/2 memory exhaustion; HTTP/2 NULL dereference on cancelled file download); CVE-2026-123867 4.2 (CVSS 3.1; per-host mTLS CA bypass). All CNA secondary; NVD “Awaiting Analysis”, no NVD assessment. |
| The three Low (CVSS, CNA) | CVE-2026-123829 2.5 (CVSS 3.0; Permission Model recursive-watch name leak); CVE-2026-123862 2.9 (CVSS 3.0; zlib params callback after destroy). CVE-2026-123873 has no NVD record — see the callout above. |
| CVE-2026-106071 (wildcard --allow-fs-read over-grant) | 3.6 Low (CVSS 3.0, CNA secondary; published 23 Oct, last modified 2 Nov; NVD lists 26.x only, references the October releases). Not in the November release notes’ 16-CVE list; see the callout above. |
| Cisco advisory v1.12 (12 Nov) | Updated products under investigation, vulnerable products and products confirmed not vulnerable. Priority products ISE, IOS XE, IOS XR, Catalyst SD-WAN, ASA, FMC and FTD are Not impacted; NX-OS Guest Shell fixes listed (Guestshell 4.9, NX-OS releases Dec 2026–Feb 2027). Still covers the October CVEs (Brotli etc.) only — no November-16-CVE advisory yet. |
| Splunk advisory SVD-2026-1015 | No new revision in the window — last update 9 Nov, documentation-only. |
| Apple HomeView Software 27.2 (12 Nov) | Lists the Brotli CVE-2026-111497 fix (out-of-bounds write, improved input validation) — the Brotli fix now reaches a new Apple product line. The older-branch gap (iOS/iPadOS 18.7.12, watchOS 26.6.1 not listing the fix) remains unexplained by Apple. |
| Confirmed production compromise | None named to 13 November. This row is empty on purpose — do not add attempts + KEV into a victim count. |
Sources: NVD — CVE-2026-123828 (5.9) · NVD — CVE-2026-123867 (4.2) · NVD — CVE-2026-123868 (5.9) · NVD — CVE-2026-123871 (6.2) · NVD — CVE-2026-123875 (5.9) · NVD — CVE-2026-123877 (6.8) · NVD — CVE-2026-123829 (2.5) · NVD — CVE-2026-123862 (2.9) · NVD — CVE-2026-106071 (3.6) · Cisco — Vulnerabilities in Node.js and Brotli Affecting Cisco Products: October 2026, v1.12, 12 Nov 2026 · 9to5Mac — Apple HomeView gets its first security update, 12 Nov 2026 (HomeView Software 27.2 lists CVE-2026-111497, quoting Apple’s advisory)
The CVE records for the 11 November batch now carry CNA scores: all seven High CVEs have CVSS 3.0 base scores from the OpenJS Foundation CNA, published to NVD on 11 November. The highest is CVE-2026-123863 at 8.2; CVE-2026-123864 and CVE-2026-123872 score 8.1; CVE-2026-123865, CVE-2026-123866, CVE-2026-123874 and CVE-2026-126051 score 7.5. Node.js itself still publishes no scores — the advisory carries severity words only — and NVD has not yet provided its own assessments. The six Medium and three Low CVEs have no score verified in this check. This supersedes the 12 November section’s “no CVSS scores published yet”; the patching advice is unchanged, and the advisory still carries no exploitation statement, with no PoC or KEV listing found for the batch in this check.
These are the CNA’s CVSS 3.0 scores, not NVD’s: NVD lists its own assessment as not yet provided on each record. All seven stay High, matching Node.js’s severity words, so the scores sharpen prioritisation inside the batch rather than changing it — the TLS private-key disclosure (CVE-2026-123864, 8.1) and the client-certificate bypass (CVE-2026-123872, 8.1) still matter most for mTLS deployments, and the 8.2 is a remotely triggerable HTTP/2 abort. Do not read a 7.5 as low priority: six of the seven Highs need no authentication at all, and the seventh (CVE-2026-123872) needs only a client certificate for a permitted host.
Sources: NVD — CVE-2026-123863 (8.2) · NVD — CVE-2026-123864 (8.1) · NVD — CVE-2026-123872 (8.1)
| Evidence | Status to 12 Nov (evening) |
|---|---|
| CVE-2026-123863 (HTTP/2 session-teardown abort) | 8.2 (CVSS 3.0, CNA openjs; AV:N/AC:L/PR:N/UI:N/S:U/C:N/I:L/A:H). Highest in the batch. |
| CVE-2026-123864 (HTTPS heap-memory disclosure, incl. TLS private keys) | 8.1 (CVSS 3.0, CNA openjs; AV:N/AC:H/PR:N/UI:N/S:C/C:L/I:L/A:H). |
| CVE-2026-123872 (TLS 1.2 renegotiation / resumption SNI change, mTLS bypass) | 8.1 (CVSS 3.0, CNA openjs; AV:N/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:N). |
| CVE-2026-123865, CVE-2026-123866, CVE-2026-123874, CVE-2026-126051 | 7.5 each (CVSS 3.0, CNA openjs): 123865 HTTP/2 session and file-descriptor leak; 123866 HTTP/2 freed response memory; 123874 respondWithFile use-after-free; 126051 TLS deferred-callback crash. |
| The six Medium and three Low | No score verified in this check. Node.js severity words (Medium / Low) stand; NVD assessment not yet provided on the records read. |
| Cisco advisory v1.11 / Splunk SVD-2026-1015 | No newer version in the window — unchanged. |
| Confirmed production compromise | None named to 12 November. This row is empty on purpose — do not add attempts + KEV into a victim count. |
Sources: NVD — CVE-2026-123863 · NVD — CVE-2026-123864 · NVD — CVE-2026-123872 · NVD — CVE-2026-123865 · NVD — CVE-2026-123866 · NVD — CVE-2026-123874 · NVD — CVE-2026-126051
Node.js shipped the security releases pre-alerted for 11 November: v26.13.1 (Lithium LTS), v24.22.3 (Krypton LTS) and v22.23.5 (Jod LTS), all dated 11 November — the releases the 5 November section announced and the 11 November sections were still waiting on. Sixteen CVEs: seven High, six Medium, three Low; highest severity High on all three lines; no CVSS scores published yet. The batch is a separate, later set of flaws — HTTP/2 and TLS, plus proxy-credential, Permission Model and zlib issues — and does not touch CVE-2026-111497, fixed on 21 October. Dependency updates: llhttp 9.4.4 (all lines), undici 8.12.4 (26.x) / 7.30.2 (24.x), sqlite 3.54.1 (22.x/24.x) and brotli 1.2.1 (24.x/26.x). Apple also closed the older-branch gap left open in the 11 November section: five of the seven same-day 9 November releases list CVE-2026-111497; iOS/iPadOS 18.7.12 and watchOS 26.6.1 do not. Cisco v1.11 and Splunk SVD-2026-1015 are unchanged in the window.
The 11 November releases fix sixteen different flaws; CVE-2026-111497 is not among them. Do not read this release as a Brotli re-fix, or as evidence the October Brotli fix failed: the Brotli update in this batch (1.2.1) is the same version that already carried the October fix. All seven High flaws are HTTP/2 or TLS, and two are client-certificate authentication bypasses — patch on the release as a whole, not on CVE-2026-111497. No CVSS scores are published with the release yet, and the advisory carries no exploitation statement; severity is Node.js’s own High / Medium / Low until scores publish.
Sources: Node.js, Wednesday, November 11, 2026 Security Releases · Node.js v26.13.1 release notes, 11 Nov 2026 · Node.js v24.22.3 release notes, 11 Nov 2026
| Evidence | Status to 12 Nov |
|---|---|
| November security releases (11 Nov) | Shipped: v26.13.1, v24.22.3, v22.23.5. Sixteen CVEs — 7 High, 6 Medium, 3 Low; highest severity High on all three lines; no CVSS scores published yet; no exploitation statement in the advisory. |
| The seven High | CVE-2026-123863 (HTTP/2 session-teardown abort, 24.x/26.x), CVE-2026-123864 (HTTPS heap-memory disclosure including TLS private keys, all lines), CVE-2026-123865 (HTTP/2 session and file-descriptor leak, all lines), CVE-2026-123866 (HTTP/2 freed response memory leaked to a peer, all lines), CVE-2026-126051 (TLS deferred-callback crash, 24.x/26.x), CVE-2026-123872 (TLS 1.2 renegotiation and resumption changing SNI, client-certificate auth bypass, all lines), CVE-2026-123874 (HTTP/2 respondWithFile use-after-free, 24.x/26.x). |
| The six Medium and three Low | Medium: CVE-2026-123828 (HTTP/2 Alt-Svc overflow), CVE-2026-123867 (per-host mTLS CA bypass), CVE-2026-123868 (HTTP/2 memory exhaustion), CVE-2026-123871 (proxy credentials in error messages), CVE-2026-123875 (HTTP/2 NULL dereference on cancelled file download), CVE-2026-123877 (https.Agent TLS policy session-cache bypass). Low: CVE-2026-123829 (Permission Model recursive-watch name leak), CVE-2026-123862 (zlib params callback after destroy), CVE-2026-123873 (Permission Model heap-snapshot destination check). |
| CVE-2026-111497 in this batch | Untouched — fixed 21 Oct in 22.23.4 / 24.22.1 / 26.12.1. Brotli updated to 1.2.1 again; no new Brotli version ships in this batch. |
| Apple older-branch verification (9 Nov releases, verified 12 Nov) | Lists CVE-2026-111497: iOS/iPadOS 26.7.3, macOS Tahoe 26.7.3, macOS Sequoia 15.8.3, watchOS 27.2.1, visionOS 27.2.1. Does not list it: iOS/iPadOS 18.7.12, watchOS 26.6.1. The gap from the 11 November section is closed. |
| Cisco advisory v1.11 (10 Nov) | No newer version in the window — 32 products with named fixed releases, unchanged. |
| Splunk advisory SVD-2026-1015 | No new revision in the window — last update 9 Nov, documentation-only. |
| Confirmed production compromise | None named to 12 November. This row is empty on purpose — do not add attempts + KEV into a victim count. |
--allow-fs-read over-grant (fixed in the release, per the commit “fix the fs permission tree node on wildcard nodes”) with no CVE attached, while the release notes list CVE-2026-123873 (Low, Permission Model heap-snapshot destination check) with no matching advisory description. Both are Low; neither changes the patching advice.Sources: Node.js, Wednesday, November 11, 2026 Security Releases · Node.js v26.13.1 release notes, 11 Nov 2026 · Apple, About the security content of iOS 26.7.3 and iPadOS 26.7.3, 9 Nov 2026 · Apple, About the security content of iOS 18.7.12 and iPadOS 18.7.12, 9 Nov 2026 · Apple, Apple security releases (9 Nov 2026 releases)
Apple released iOS 27.2.1 and iPadOS 27.2.1 and macOS Golden Gate 27.2.1 on Monday 9 November, and each advisory lists CVE-2026-111497 — “Processing maliciously crafted input may lead to memory corruption”, fixed with improved input validation. The iOS release covers iPhone 11 and later and the listed iPad models. Apple does not flag the CVE — or any of the nine entries in the iOS release — as exploited. This is a new fixed product family for the same CVE: the page’s vendor-remediation record (Node.js, .NET, Python, Bun, PHP for Windows, Cisco, Splunk) now extends to Apple devices. The releases are dated 9 November and were missed by the 10 and 11 November morning checks; they are recorded here with their true date. The 11 November Node.js security releases still have not landed — the pre-alert is unchanged, with no CVE IDs, components or fixed versions published.
Apple’s iOS 27.2.1 advisory lists CVE-2026-111497 alongside eight other fixes; none of the nine is flagged as exploited. Do not read the patch as evidence of in-the-wild exploitation on Apple devices. Apple’s advisory names the CVE and the impact; it does not say which component carries the decoder or how it is reached on the device — treat the listing as a patch confirmation, not an exposure assessment.
Sources: Apple, About the security content of iOS 27.2.1 and iPadOS 27.2.1, 9 Nov 2026 · Apple, About the security content of macOS Golden Gate 27.2.1, 9 Nov 2026 · Node.js, November 2026 security releases pre-alert, unchanged to 11 Nov
| Evidence | Status to 11 Nov |
|---|---|
| Apple iOS / iPadOS 27.2.1 (9 Nov) | CVE-2026-111497 listed: out-of-bounds write, fixed with improved input validation. iPhone 11 and later, listed iPad models. Not flagged as exploited. |
| Apple macOS Golden Gate 27.2.1 (9 Nov) | Same CVE-2026-111497 listing, same impact and fix wording. Not flagged as exploited. |
| November security release pre-alert | Unchanged to 11 Nov — releases for 26.x, 24.x and 22.x due “on or shortly after” Wednesday 11 November; still no CVE IDs, components or fixed versions published. |
| Cisco advisory v1.11 (10 Nov) | No newer version in the window — 32 products with named fixed releases, unchanged. |
| Splunk advisory SVD-2026-1015 | No new revision in the window — last update 9 Nov, documentation-only. |
| Confirmed production compromise | None named to 11 November. This row is empty on purpose — do not add attempts + KEV into a victim count. |
Sources: Apple, About the security content of iOS 27.2.1 and iPadOS 27.2.1, 9 Nov 2026 · Apple, About the security content of macOS Golden Gate 27.2.1, 9 Nov 2026 · Apple, Apple security releases (9 Nov 2026 releases)
Cisco updated its Node.js advisory to v1.11 on Tuesday 10 November (21:02 UTC): 32 products now carry named fixed releases (was 28 at v1.9 on 9 November), and the six priority products — Catalyst SD-WAN, IOS XE, IOS XR, Secure Firewall ASA, FMC and FTD — remain confirmed not impacted, with ISE still under investigation. Splunk’s advisory SVD-2026-1015 has no new revision in the window. The 11 November Node.js security releases have not landed yet — the pre-alert is unchanged, with no CVE IDs, components or fixed versions published — and the KEV remediation date for CVE-2026-111497 passed on 10 November with no new development attached to it.
Both advisories cover the four October Node.js CVEs as a group — Brotli CVE-2026-111497 plus the DNS and TLS flaws CVE-2026-105482, CVE-2026-105483 and CVE-2026-106069 — so a product on either list is not necessarily Brotli-affected. Check the per-product detail before treating a listing as a Brotli exposure. The Cisco count is now from Cisco’s own advisory (v1.11, 10 Nov); the Splunk list is from Splunk’s own advisory, linked below.
Sources: Cisco, Vulnerabilities in Node.js and Brotli Affecting Cisco Products, v1.11, 10 Nov 2026 · Node.js, November 2026 security releases pre-alert, unchanged to 11 Nov
| Evidence | Status to 11 Nov |
|---|---|
| Cisco advisory v1.11 (10 Nov) | 32 products with named fixed releases (was 28). Catalyst SD-WAN, IOS XE, IOS XR, Secure Firewall ASA, FMC and FTD confirmed not impacted; ISE still under investigation. PSIRT still aware of no malicious use of these vulnerabilities against any Cisco product. |
| Splunk advisory SVD-2026-1015 | No new revision in the window — last update 9 Nov, documentation-only. 13 products with affected packages, unchanged. |
| November security release pre-alert | Unchanged to 11 Nov — releases for 26.x, 24.x and 22.x due “on or shortly after” Wednesday 11 November; still no CVE IDs, components or fixed versions published. |
| Confirmed production compromise | None named to 11 November. This row is empty on purpose — do not add attempts + KEV into a victim count. |
Sources: Cisco, Vulnerabilities in Node.js and Brotli Affecting Cisco Products, v1.11, 10 Nov 2026 · Node.js, November 2026 security releases pre-alert, unchanged to 11 Nov
Cisco updated its Node.js advisory to v1.9 on Monday 9 November: 28 products now carry named fixed releases (was 25 at v1.8 on 6 November), and the six priority products — Catalyst SD-WAN, IOS XE, IOS XR, Secure Firewall ASA, FMC and FTD — remain confirmed not impacted, with ISE still under investigation. Splunk’s advisory SVD-2026-1015, also updated 9 November, changed only the Visual Exporter mitigation wording — no product-status change. No confirmed production compromise, victim or new IOC set was found to 10 November, and the 11 November pre-alert is unchanged; the KEV remediation date for CVE-2026-111497 falls today, with no new development attached to it.
Both advisories cover the four October Node.js CVEs as a group — Brotli CVE-2026-111497 plus the DNS and TLS flaws CVE-2026-105482, CVE-2026-105483 and CVE-2026-106069 — so a product on either list is not necessarily Brotli-affected. Check the per-product detail before treating a listing as a Brotli exposure. The Cisco count is now from Cisco’s own advisory (v1.9, 9 Nov); the Splunk list is from Splunk’s own advisory, linked below.
Sources: Cisco, Vulnerabilities in Node.js and Brotli Affecting Cisco Products, v1.9, 9 Nov 2026 · Splunk, SVD-2026-1015, updated 9 Nov 2026
| Evidence | Status to 10 Nov |
|---|---|
| Cisco advisory v1.9 (9 Nov) | 28 products with named fixed releases (was 25). Catalyst SD-WAN, IOS XE, IOS XR, Secure Firewall ASA, FMC and FTD confirmed not impacted; ISE still under investigation. PSIRT still aware of no malicious use of these vulnerabilities against any Cisco product. |
| Splunk advisory SVD-2026-1015 (updated 9 Nov) | Documentation-only change — the Visual Exporter mitigation now names the functionality it may break. 13 products with affected packages, unchanged; Galileo still remedied; Splunk Synthetic Runner fix still 1.71.0 or higher. |
| November security release pre-alert | Unchanged to 10 Nov — still no CVE IDs, components or fixed versions published for the 11 November release. |
| Confirmed production compromise | None named to 10 November. This row is empty on purpose — do not add attempts + KEV into a victim count. |
Sources: Cisco, Vulnerabilities in Node.js and Brotli Affecting Cisco Products, v1.9, 9 Nov 2026 · Splunk, SVD-2026-1015, updated 9 Nov 2026
Cisco updated its Node.js advisory to v1.8 on Friday 6 November: 25 products now carry named fixed releases (was 19), and six priority products — Catalyst SD-WAN, IOS XE, IOS XR, Secure Firewall ASA, FMC and FTD — are confirmed not impacted, with ISE still under investigation. Splunk’s advisory SVD-2026-1015, also updated 6 November, lists 13 products with affected packages (was 11); Galileo is now remedied and Splunk Synthetic Runner has a fix (upgrade to 1.71.0 or higher). No confirmed production compromise, victim or new IOC set was found to 7 November, and the 11 November pre-alert is unchanged.
Both advisories cover the four October Node.js CVEs as a group — Brotli CVE-2026-111497 plus the DNS and TLS flaws CVE-2026-105482, CVE-2026-105483 and CVE-2026-106069 — so a product on either list is not necessarily Brotli-affected. Check the per-product detail before treating a listing as a Brotli exposure. The Cisco count is now from Cisco’s own advisory (v1.8, 6 Nov); the Splunk list is from Splunk’s own advisory, linked below.
Sources: Cisco, Vulnerabilities in Node.js and Brotli Affecting Cisco Products, v1.8, 6 Nov 2026 · Splunk, SVD-2026-1015, updated 6 Nov 2026
| Evidence | Status to 7 Nov |
|---|---|
| Cisco advisory v1.8 (6 Nov) | 25 products with named fixed releases (was 19). Catalyst SD-WAN, IOS XE, IOS XR, Secure Firewall ASA, FMC and FTD confirmed not impacted; ISE still under investigation. Some fixes are months out — Catalyst Center 3.4.1 in July 2027, Smart PHY in March 2027. |
| Splunk advisory SVD-2026-1015 (updated 6 Nov) | 13 products with affected packages (was 11). Galileo now remedied; Splunk Synthetic Runner has a fix — upgrade to 1.71.0 or higher. Splunk Enterprise 10.0 guidance unchanged (upgrade to 10.2 or higher); Visual Exporter still disable or remove in all Enterprise versions. |
| November security release pre-alert | Unchanged to 7 Nov — still no CVE IDs, components or fixed versions published for the 11 November release. |
| Confirmed production compromise | None named to 7 November. This row is empty on purpose — do not add attempts + KEV into a victim count. |
Sources: Cisco, Vulnerabilities in Node.js and Brotli Affecting Cisco Products, v1.8, 6 Nov 2026 · Splunk, SVD-2026-1015, updated 6 Nov 2026
Cisco updated its Node.js advisory on Wednesday 4 November (reported 5 November): 19 Cisco products affected, dozens more still under investigation. Splunk’s advisory SVD-2026-1015, updated 5 November, lists 11 products with affected packages. No confirmed production compromise, victim or new IOC set was found to 6 November, and the 11 November pre-alert is unchanged.
Both advisories cover the four October Node.js CVEs as a group — Brotli CVE-2026-111497 plus the DNS and TLS flaws CVE-2026-105482, CVE-2026-105483 and CVE-2026-106069 — so a product on either list is not necessarily Brotli-affected. Check the per-product detail before treating a listing as a Brotli exposure. The Cisco count comes from SecurityWeek’s report of the updated advisory; the Splunk list is from Splunk’s own advisory, linked below.
Sources: SecurityWeek, Cisco, WatchGuard release patches, 5 Nov 2026 · Splunk, SVD-2026-1015, updated 5 Nov 2026
| Evidence | Status to 6 Nov |
|---|---|
| Cisco advisory update (4 Nov, reported 5 Nov) | 19 Cisco products affected; dozens more under investigation. Fixes rolling per product. |
| Splunk advisory SVD-2026-1015 (updated 5 Nov) | 11 products with affected packages, including Splunk Enterprise 10.0 (upgrade to 10.2 or higher) and the Splunk Visual Exporter app, affected in all Enterprise versions — disable or remove it as mitigation. Galileo still under investigation. |
| November security release pre-alert | Unchanged to 6 Nov — still no CVE IDs, components or fixed versions published for the 11 November release. |
| Confirmed production compromise | None named to 6 November. This row is empty on purpose — do not add attempts + KEV into a victim count. |
Sources: SecurityWeek, Cisco, WatchGuard release patches, 5 Nov 2026 · Splunk, SVD-2026-1015, updated 5 Nov 2026
Node.js has announced security releases for its 22.x, 24.x and 26.x lines on or shortly after Wednesday 11 November, the highest-severity fix in each line rated High. No CVE IDs, components or fixed versions are published yet — and no confirmed production compromise, victim or new IOC set was found to 5 November.
CISA KEV remediation for CVE-2026-111497 falls due 10 November — the day before the new releases are expected. The November release is a separate batch: the KEV fix is the October release (22.23.4 / 24.22.1 / 26.12.1), not this one. Do not read the 11 November date as an extension of the KEV deadline.
Source: Node.js, Wednesday, November 11, 2026 Security Releases
| Evidence | Status to 5 Nov |
|---|---|
| November security release pre-alert | Announced — Node.js, 3 Nov; oss-security pre-alert 4 Nov. New versions of 22.x, 24.x and 26.x on or shortly after 11 Nov. |
| Highest severity in the batch | High in each line. No CVE IDs, components or fixed versions yet; details embargoed until release. |
| CISA KEV deadline (CVE-2026-111497) | Still pending, due 10 Nov — the day before the new release. Unchanged from the 4 November update. |
| Confirmed production compromise | None named to 5 November. This row is empty on purpose — do not add attempts + KEV into a victim count. |
Sources: Node.js, Wednesday, November 11, 2026 Security Releases · oss-security, Node.js security release pre-alert, 4 Nov 2026
The picture did not harden further. No confirmed production compromise, no new named victim, no new public IP IOC set and no named threat actor were found to 4 November — and the KEV deadline that matters next had not yet passed.
CISA KEV added 27 October, federal remediation due 10 November, ransomware use “unknown”. At the 4 November audit the deadline was 6 days away; show it as a countdown, not as met or missed.
Sources: CSA AL-2026-154 summary
| Evidence | Status to 4 Nov |
|---|---|
| CISA KEV listing (27 Oct) | Yes — KEV means exploitation in the wild met the listing threshold, not a named breach. |
| Decoy / probing attempts | Zscaler: 402 decoy events, 20 IPs, 12 customers in the first 24 hours after disclosure (40% recon, 60% exploit attempts). Check Point telemetry below is probing, not compromise. |
| Lab RCE | Disputed: lab RCE on unmodified Node.js 24.22.0 via Express 5 exists, with low per-attempt success; Red Hat says it knows of no reliable public PoC for code execution. Unresolved. |
| Confirmed production compromise | None named to 4 November. This row is empty on purpose — do not add attempts + KEV into a victim count. |
EPSS moves daily and sources differ; a chart without dates would mislead. 27 Oct values differ by source (FIRST 55% vs Tenable FAQ 36%); treat that as a source/method difference, not a daily move.
Check Point explicitly states this does not confirm successful compromise, and the percentages are its telemetry, not global attack volume. The new behavioural addition is the Patrowl fingerprint, not a new IP list.
11.2.4-h23, 12.1.4-h13, 12.1.10-h3); earlier branches listed 29 Oct are not repeated here; no PAN-OS exploitation known. Scope is narrow: PA/VM-Series only when Brotli inspection and TLS/SSL decryption are both on (Brotli off by default); management UI/API not affected. Datadog confirms in-the-wild exploitation and adds detections, with no new victim named in the summaries found.Sources: Check Point, HTTP2Shell · Arista advisory 0202 · Mallory, Node2Shell rapid response
The picture hardened without changing: exploitation is now officially confirmed, telemetry has produced the first address list, and nothing in the window shows a successful shell against a production target.
Reported by SecurityWeek, alongside Previdian's record of one attempt on 27 October from a US-based IP (not published), with the explicit qualification that the attempt did not succeed. Broadcom's advisory makes no in-the-wild statement. The Canadian Cyber Centre's AL26-027 (28 Oct) carries zero IOCs and no compromise statement.
So the standing grading holds: the vulnerability is certain, code execution is contested-but-demonstrated in labs, and in-the-wild exploitation is now confirmed as attempts with no successful RCE and no victim.
Sources: SecurityWeek, 28 Oct · Canadian Cyber Centre AL26-027
Deception telemetry from the first 24 hours: 20 source IPs, 12 customers, 402 events, split 40% reconnaissance / 60% exploitation. The five exploitation addresses — 49.37.46.99, 34.10.240.142, 188.245.184.87, 52.46.50.66, 136.111.197.96 — are now in iocs/iocs.csv at MONITOR. No payload, outcome or victim detail accompanies them.
The durable part is the fingerprints, not the addresses: t13d251100_b78ed14e2fd0_ab7e3b40a677 for the 22/23-byte invalid block-switch seeds, t13d4312h1_c7886603b240_b26ce05bbdd6 for malformed-Brotli plus Node-path recon, and t13d1714h1_ab0a1bf427ad_5e078a691127 for the 14-byte {"ping":1} sweep. Address lists rotate; a JA4 that encodes the trigger does not.
Source: Zscaler ThreatLabz, 28 Oct
v24.22.2 and v26.13.0 on 28 October — neither carries Brotli content (V8 backports; the 26.x release is an LTS transition). .NET unchanged since 25 Oct. No in-window release for ClickHouse, DuckDB, Arrow, libjxl, 7-Zip-zstd, YDB, woff2 or Bun. Debian nodejs remains vulnerable in bookworm, trixie and forky — fixed only in sid.Three things changed since 26 October. One invalidates a headline in this brief (“no KEV entry”), one narrows another (“no network IOCs”), and one fixes a carrier table that was already stale when it was published.
2026.10.27, 1,760 entries
CVE-2026-111497 is listed. Vendor Google, product Brotli, name “Google Brotli Unchecked Return Value Vulnerability”, added 27 October, dueDate 2026-11-10, CWEs 252 and 787, knownRansomwareCampaignUse: Unknown, forensicTriage: No. Fetched directly from the KEV JSON feed on 28 October and announced in the 27 October alert.
The machine-readable reason is in the NVD record — a CISA Coordinator SSVC block timestamped 27 October 19:53 UTC with exploitation: active, automatable: no, technicalImpact: total. That reverses the exploitation: none this brief quoted from 23 October. EPSS followed: 0.55483, 99.03rd percentile on 27 October, up from 0.3278.
KEV does not mean a shell was obtained. It records that exploitation was observed. The single observed attempt, below, failed — and a failed attempt against a memory-corruption bug is still a remote crash primitive, which is why the listing is defensible on the evidence available.
94.154.43.7ELLIO recorded exploit traffic on 26 October, one day after its own reconnaissance-scanner observation. The detail is unusually complete, and it is the first time this case can describe an actual attack rather than a scanner.
| Attribute | Observed |
|---|---|
| Time | 26 Oct 2026, 15:08 UTC |
| Source | 94.154.43.7 — first observed 15 Dec 2025, active 47 of the last 90 days, 25 behaviour tags, six CVEs on record, previously seen on PHP2Shell, GlobalProtect and FortiOS SSL-VPN probing |
| Request body | 23,069,983 bytes, SHA-256 7fcbdaaecd4419e2930927844f3b772668204737a4e7602a2290c991c93c5c4b |
| Structure | Two stages: a 1,031-byte instruction block and a 256-byte stage two. The first 23,069,727 bytes reproduce the public stream generator byte for byte; the ROP chain is original — the third distinct construction observed |
| Target | One specific binary: official Node.js v24.22.0, x86-64. Every address in the chain presumes it |
| Result | Failed. ELLIO's test of the attacker's chain: zero hits in 50 attempts against a pristine target. The pivot works and the process crashes, but it never reaches execvp. Their description: “a working framework with a defective last 12 instructions” |
| Difference from a working chain | The attacker's chain redirects stdin only, so commands go in and nothing comes out. ELLIO's own chain redirects both, which is why its recordings show id and uname -a answering back |
Why this matters more than the KEV label: it shows a real actor building on the public PoC with an original final stage and getting it wrong. That is what first-generation exploitation of a hard memory-corruption bug looks like, and it is the most likely shape of the next attempt too — which is also why crash telemetry is the durable detection, not shell telemetry.
iocs/iocs.csv has been updated from header-only to two rows, and iocs/INDICATORS.md now carries the correction rather than the original claim.HttpClient automatic decompression are off by default, and .NET Framework is unaffected.ALAS2023-2026-3334 (important): brotli, published 24 October, plus ALAS2023-2026-3330 for nodejs24. Verified in the ALAS RSS feed on 28 October.The tail this brief called “still unfixed” split in three directions. Full tracker: broccoli2shell.com, updated 27 October.
| State | Products |
|---|---|
| Fixed in a release | .NET 8.0.33 / 9.0.22 / 10.0.14 / 11.0.0-rc.3 (25 Oct); Amazon Linux 2023; Red Hat RHEL 8/9/10 via RHSA-2026:90501 / 90495 / 90449 (26 Oct); Oracle Linux ELSA-2026-500514; AlmaLinux; Debian DSA-6582-1 and DLA-4855-1; Ubuntu (23 Oct, updated 26 Oct); Alpine 3.21/3.23/edge; NanaZip 7.0.1882.0; Ruby gem 0.9.1; PHP brotli ext 0.21.1; UEFITool A76; MuPDF 1.28.0; Ghostscript 10.07.1rc1; Go andybalholm/brotli v1.2.7 (tag only, no release object) |
| Fix merged, not released | ClickHouse (PR #127419, backported to 26.3/26.8/26.9/26.10 LTS — latest release pre-dates it); libjxl (PR #5062); GraalJS; YDB |
| Still unfixed | DuckDB (PR #27791 open); Apache Arrow (bundles 1.0.9; PR #55167 approved, blocked on wheels); 7-Zip-zstd; p7zip; EDK II; woff2 Makefile build; Debian nodejs in bookworm-security (18.20.4) and trixie-security (20.19.2) — only sid is fixed; Pillow, pyarrow, curl-cffi, pycurl wheels that vendor their own decoder |
PAN-OS is affected where Brotli inspection and TLS/SSL decryption are both enabled on the dataplane — no management-interface impact, and Brotli inspection does not exist before 11.2.4, so 11.1 and 10.2 are out of scope. Fixed in 12.2.4-h1, 12.1.7-h11, 11.2.10-h17; several branches were still on ETA between 27 and 30 October.
Hotfix trap: 11.2.10-h16 — the hotfix that fixes PHP2Shell — does not fix this one. 11.2.10-h17 covers both. On the same product, the PHP CVEs hit the management plane and this one hits the dataplane; a team that applied h16 for PHP2Shell will believe it is patched and will not be.
Two more carriers worth adding to inventory: git-remote-https wherever libcurl is built with brotli (true on most distributions), and FFmpeg ≥8.1 with --enable-libjxl, where libjxl decodes Brotli-compressed Exif metadata in a crafted .jxl. Also brotli4j <1.24.0 for Java.
Neither position has been refuted, and the honest reading is that they are now answering different questions.
node:http, official node:24.22.0-slim as root, Express 5.3.0, Koa 3), default configuration, ASLR left on, at roughly attempt 5. That is broader than Imperva's arm64-only result.AC:High rather than Critical — citing ASLR, NX, FORTIFY and SELinux. That remains true: ELLIO and Imperva have both withheld exploit code.Content-Encoding: br request shapes. Blevene’s gist adds 8 Sigma rules and Suricata sid 2026114971 (inbound br token) plus an egress flowbits pair. Fortinet ships IPS DB 37.321; Rapid7 added an authenticated check in its 26 Oct content release.br bodies — and treat the resulting core dumps as a data-loss event, because with systemd-coredump they land in /var/lib/systemd/coredump and can contain other users’ session tokens. Note the failure mode differs by ecosystem: Python Brotli/brotlicffi, the PHP extension’s streaming API and the Ruby gem hang instead of crashing; PHP spins at 100% CPU.process.versions.brotli as a patch check. It reads 1.2.0 on both patched and unpatched Node.js builds. Check the Node release version instead.Several vendor pages are stamped with a 27 October update date and still say the opposite of this update: Rapid7’s reads “not aware of exploitation in the wild” and “not listed in CISA’s KEV”; Tenable’s FAQ and BleepingComputer’s 26 October piece say the same. All three are superseded by the 27 October KEV entry. Check the KEV feed, not the vendor’s page date.
Read this before any of the exploitability argument, because the argument is entirely about whether this write is reachable.
Brotli's DecodeBlockTypeAndLength() must decide whether a malformed block switch is a fatal format error or a request for more input. It conflated the two: a fast path ignored the return value, so the streaming decoder treated an invalid block switch as "need more bytes" and kept appending into an eight-byte internal buffer. The overrun writes into adjacent decoder state.
The upstream fix is commit 0ed4268a — "Detect and report invalid stream earlier" — which introduces the distinct error BROTLI_DECODER_ERROR_FORMAT_BLOCK_SWITCH (-17). Related commits: 50a9b70, b418fa8, da98063.
The fix landed upstream in February 2026 and was not labelled a security fix. Disclosure followed eight months later, on 21 October, with the Node.js release. Anything that vendored, pinned or bundled a Brotli build between those dates is vulnerable regardless of when it was last "updated". A dependency manifest showing a recent build proves nothing; you have to know which Brotli that build contains.
Reported by OpenAI Codex Security — Khai Tran and Sylvie Mayer — and fixed by mcollina and aduh95. Tracked as CVE-2026-111497 / GHSA-9v8h-qqgm-4m35, CWE-787 with CWE-252 as the root cause. Nicknames in circulation: Node2Shell, HTTP2Shell, Broccoli2Shell, Python2Shell — the proliferation of names is itself a sign the blast radius is not Node.js.
The single most useful thing in this report: the flaw is redistributed inside runtimes, wheels, proxies and vendored binaries. Node.js is the loudest carrier, not the only one.
| Product / ecosystem | Vulnerable | Fixed |
|---|---|---|
| Google Brotli (C reference) | 1.0.3–1.0.9, 1.1.0, 1.2.0 | 1.0.10, 1.1.1, 1.2.1 |
| Node.js 22 / 24 / 26 | <22.23.4 / <24.22.1 / <26.12.1 | 22.23.4, 24.22.1, 26.12.1 |
| Node.js 18 / 20 / 23 / 25 | all lines | none — EOL |
| Bun | 1.0.21–1.4.3 | 1.4.4 |
PyPI Brotli | 1.0.4, 1.0.6–1.0.9, 1.1.0, 1.2.0 | 1.2.1 |
PyPI brotlicffi | 1.0.9.0–1.0.9.2, 1.1.0.0, 1.2.0.0–1.2.0.2 | 1.2.1.0 |
| Envoy | 1.39.2, 1.38.5, 1.37.7, 1.36.11 and earlier | 1.40.0 / 1.39.3 / 1.38.6 / 1.37.8 / 1.36.12 |
| Istio | 1.31.0–1.31.1, 1.30.0–1.30.5, 1.29.0–1.29.8 | 1.31.2, 1.30.6, 1.29.9 |
| cdxgen (standalone) | ≤13.4.0, ≤12.8.8 | 13.5.0, 12.8.9 |
| Debian bookworm / trixie | libbrotli | 1.0.9-2+deb12u1 / 1.1.0-2+deb13u1 |
| Ubuntu 24.04 / Alpine 3.24 | libbrotli1 / brotli | 1.1.0-2ubuntu0.1 / 1.2.1-r0 |
Go andybalholm/brotli | before 23 Oct fix | fixed — crash/panic only |
| Still unfixed at publication | .NET runtime (bundles 1.1.0; hits BrotliDecoder, BrotliStream, HttpClient auto-decompression), DuckDB, ClickHouse, Apache Arrow (1.0.9), libjxl, NanaZip, 7-Zip-zstd, p7zip, EDK II firmware, GraalJS, YDB, Debian nodejs, Amazon Linux 2023, woff2 built from a Makefile | |
Also reported fixed: Chromium, Firefox, ngx_brotli, Brotli4j, Ghostscript, SumatraPDF, Ruby gem, PHP extension, UEFITool, MuPDF. Electron ships both a Chromium and a Node copy. Pure Java/JS/C# ports (pdf.js, iText, HtmlUnit) are stated not affected.
You do not need to import Brotli to be exposed. requests, urllib3, aiohttp, httpx and Scrapy auto-decode Content-Encoding: br when either Brotli package is installed. Worse, earlier urllib3 guidance told users to require Brotli ≥1.2.0 or brotlicffi ≥1.2.0.0 — which are the last vulnerable releases. Extras such as urllib3[brotli], httpx[brotli] and aiohttp[speedups] pull them in on purpose.
The name changed twice, and each change was a correction of scope rather than branding. The final name is the one that should drive your inventory search.
| Name | Date | Claimed scope | What it added |
|---|---|---|---|
| Node2Shell | 21 Oct | Node.js | Node 22/24/26, Bun, and anything vendoring the decoder through a Node runtime |
| HTTP2Shell | 23 Oct Aikido headline |
Whatever decompresses HTTP | Python (Brotli, brotlicffi, and everything that pulls them in), curl/libcurl, native libbrotli, Envoy, Istio, OS packages, C++ consumers. Aikido also notes “another popular term: Broccoli2Shell” |
| Brotli2Shell | 25 Oct JFrog |
The decoder, wherever it lives | JPEG XL via FFmpeg, .NET compression and HttpClient, Apache Arrow / Parquet, ClickHouse. JFrog’s argument: the bug “belongs to the decoder, not to Node.js or HTTP itself” |
Aikido’s rename widened the HTTP perimeter. JFrog then showed the perimeter is not HTTP at all:
| Surface | Demonstrated | Impact reached |
|---|---|---|
| FFmpeg → libjxl → JPEG XL | A crafted .jxl with a malformed brob box; valid outer structure with real image data, malformed metadata reaching the decoder | Instruction-pointer control, crash at 0x414141414141. RCE “would require further exploitation” and depends on build mitigations |
| .NET | BrotliStream / System.IO.Compression.Brotli via a two-stage stream; also reachable through HttpClient auto-decompression and ASP.NET Core request-decompression middleware | Native SIGSEGV |
| Apache Arrow / pyarrow | Brotli-compressed Parquet pages and the direct buffer-decompression API — no network involved | A 534-byte input caused an infinite loop and indefinite CPU consumption; Brotli 1.2.1 rejected the same input in under a second. Issue still open |
| ClickHouse | A 23 MB request body with Content-Encoding: br — HTTP, but not Node | Segfault during decoder cleanup. Fix merged but no patched release shipped at publication |
Not one of JFrog’s four additional cases is a completed RCE. They are instruction-pointer control, native crashes, and an infinite loop. That is the third independent confirmation of the grading in this report: the primitive is real and reachable in far more places than Node, but the distance from the primitive to a shell is long, target-specific, and nobody has closed it outside Node.
What they do prove is reachability — and reachability is the whole inventory problem. A 534-byte Parquet page that hangs a worker is a denial-of-service a file upload can deliver, with no HTTP stack in the path at all.
JFrog’s framing is the right one to hand to an infrastructure team, because it splits one unanswerable question into two answerable ones:
“Yes” to the first and “no” to the second is not an emergency. “Yes” to both is. Most inventories answer only the first, and a scanner match is evidence of presence, not proof of exploitability.
br from Accept-Encoding is not a fix
A widespread mitigation circulating is to stop advertising Brotli in Accept-Encoding. On the outbound path that is insufficient — an attacker-controlled server can return Content-Encoding: br regardless of what you asked for, and a client that trusts the header will decode it. You must refuse to decode Brotli responses from untrusted origins, not merely decline to request them. Treat any guidance that stops at Accept-Encoding as incomplete.
The useful corollary, from Aikido: in default configurations Apache and nginx do not decode Brotli, so neither is a vulnerable carrier at the edge. The exposure is in the application and its client libraries, which is where inventories usually stop looking.
The naming proliferation is being mocked, and the mockery is worth knowing about because it predicts how the issue will be deprioritised. The most-shared formulation: “Three days from now → #php2shell. Today → #http2shell. One week from now → #server2shell. One year from now → #human2shell.” Another calls it a cyber soap opera that “drowns devs in an avalanche of buzzwords”.
Both things are true. The name inflation is real and mildly absurd. But the sequence underneath it is a factual widening: each name was coined because someone demonstrated the decoder somewhere the previous name did not cover, and the last one came with four reproductions. The practical reading: ignore the marketing, act on the second question above — whether untrusted input reaches a vulnerable decoder in your estate, whatever the stack is called this week.
If you are tracking remediation across language ecosystems, one CVE has several records: GHSA-9v8h-qqgm-4m35 (general), PYSEC-2026-4322 (PyPI Brotli), PYSEC-2026-4323 (brotlicffi), and GHSA-3wvj-j6qm-62mr (Envoy). Dependency scanners keyed only to the CVE ID may miss the Python and Envoy entries.
Three shapes. The inbound one is the one WAFs see; the outbound one they never will.
express.json() and raw() inflate Brotli by default (inflate: true). This is the widest exposure in the ecosystem.br.@koa/bodyparser, koa-body, co-body; also Egg and Strapi.@fastify/compress.A bare node:http server does not decode request bodies on its own. Express and body-parser are routes into the decoder, not the source of the flaw — which is why patching the runtime beats patching the framework.
Any service that fetches a user-supplied URL and auto-decompresses the reply is exposed: Node fetch()/undici/axios/got, and Python requests/urllib3/httpx/aiohttp/Scrapy. The relevant features are the ones that fetch arbitrary URLs by design — webhooks, link previews, URL imports, image proxies, OIDC discovery, crawlers, and browsing AI agents.
Imperva states its rules cover inbound only. A server-side fetch() that pulls a hostile Brotli response is egress traffic your edge never inspects and never compressed on the way in. Teams that "added a WAF rule" and stopped there have covered roughly half the exposure — and the half that is easiest to weaponise, because the attacker controls the server answering the request.
Anything that decodes Brotli inside a container format: Parquet, JPEG XL, WOFF2 fonts, some PDF paths, archives, and firmware images. This is why the "we don't serve compressed HTTP" answer is insufficient.
This is where the reporting diverges, and the divergence is substantive rather than tribal.
| Claim | Who | What they actually showed | Grade |
|---|---|---|---|
| RCE on stock build, arm64 | Imperva, 22 Oct | Execution against the official node:24.22.0 Linux arm64 image, in fresh containers, via both a fetch() response and an Express request body. “Execution took multiple attempts”; “did not succeed on every attempt”; other builds untested. PoC code not released. 24.22.1 rejected the input. |
MEDIUM |
| RCE on stock build, x86-64, four frameworks | ELLIO, ~25 Oct | Plain node:http, node:24.22.0-slim as root, Express 5.3.0 and Koa 3. Shell at ~attempt 5 with ASLR on. Needed a per-build memory offset and a predictable accepted-socket fd; most of 400 Express attempts crashed the worker instead. |
MEDIUM |
| Write is unreachable; impact is DoS | Independent researcher, via social | Argues the out-of-bounds write cannot be reached in the intended path, and that the demonstrated “RCE” required a deliberately weakened, non-PIE, ASLR-off harness rather than a stock target. | UNRESOLVED |
| No reliable public RCE PoC | Red Hat | Rates the issue Important, reports no reliable public RCE proof of concept; public repos broadly produce crashes. | MEDIUM |
How to read this. The two positive reports are not contradictory — they are on different architectures and both required repetition, which is what a heap-adjacent write under ASLR looks like. But neither released code, and both describe success as probabilistic. The sceptical position is not crankery: it correctly identifies that a crash is not a shell, and that some public demonstrations used harnesses that made success much easier.
What is not in dispute: the patched release rejects the input every time, returning 400 with ERR__ERROR_FORMAT_BLOCK_SWITCH. That is a clean, usable verification signal.
GitHub on 26 October: 30 repositories match node2shell, 52 match the CVE. First repos appeared roughly five hours after the Node.js release.
| Repository | Stars | What it is |
|---|---|---|
| Rabbitbong/node2shell | 82 | Most-starred; exploit-oriented |
| assetnote/node2shell-scanner | 57 | The one to use. High-fidelity detection from a named research team (Searchlight Cyber) |
| dimasma0305/CVE-2026-111497-POC | 46 | “Reproducible local Brotli decoder proof” — local, not remote |
| trailofbits/brotli-checker | 46 | Inventory skill for finding vulnerable Brotli copies — defensive |
| CaterMaoZi/CVE-2026-111497-POC | 15 | Chinese-language PoC |
| syrex1013, hypnguyen1209, ZeroHackOrg, shivrajem, tc4dy, TheFool-TAROT | 3–8 | Mixed PoC/lab/detector projects |
Grade the repos, not the star count. The pattern from the sibling PHP2Shell case repeats exactly: the most-starred artefact is the exploit-claiming one, the most useful artefact is a scanner from an established team, and the PoCs that "work" mostly crash the target. Two specific cautions:
There is no payload signature worth writing. The signals are version, crash, and process behaviour.
Patched builds answer malformed block switches with a specific error rather than silence:
BROTLI_DECODER_ERROR_FORMAT_BLOCK_SWITCH (-17)
Node HTTP surface: 400 + ERR__ERROR_FORMAT_BLOCK_SWITCH
Hunting for -17 on patched hosts is the cleanest way to find someone probing you. On unpatched hosts it never fires — absence proves nothing there.
Crashes can produce core dumps containing live credentials from the process heap. Containers often lack a working core handler, which is luck rather than design. If you enable core dumps for debugging, you have created a credential-exposure path on a service that a remote attacker can crash at will.
express.json({ inflate: false }) returns 415 — note it also rejects gzip, so test first. Reject Content-Encoding: br on request bodies at the proxy or load balancer.br. This is the step most teams will skip — and dropping br from Accept-Encoding does not achieve it, because a hostile server can send Content-Encoding: br anyway.-17 on patched hosts, crash-then-execute patterns, and worker restart loops correlated with compressed request bodies.Node 18, 20, 23 and 25 will never be fixed. If you are on one of those lines, the only remediation is to move.
On 26 October this section said there were no network IOCs. That was true then and is wrong now — two values exist, both from the 26 October attempt, and both corroborated by two sources.
One IP and one request-body hash have been published, and both are in iocs/iocs.csv. They describe one actor’s single failed attempt, not a campaign. Everything else in circulation is either a scanner or paywalled.
Still true, and worth keeping: one commercial intel page advertises 31 indicators and 9 detection rules, all behind a paywall — no values are disclosed, so they cannot be reproduced, graded, or cited. A social post describing “8 IOCs” for that report overstates what is visible.
| Type | Value | What it is | Sources |
|---|---|---|---|
| IPv4 | 94.154.43.7 | Source of the single observed exploit attempt, 26 Oct 15:08 UTC. Long-running multi-purpose scanner: first seen 15 Dec 2025, active 47 of the last 90 days, 25 behaviour tags, six CVEs, and prior PHP2Shell exploitation as well as GlobalProtect and FortiOS SSL-VPN probing | ELLIO, plus GreyNoise noise: true / malicious, last seen 27 Oct — two independent sources |
| SHA-256 | 7fcbdaaecd4419e2930927844f3b772668204737a4e7602a2290c991c93c5c4b | The 23,069,983-byte Brotli request body built for Node.js 24.22.0 x86-64. Public delivery layer, original ROP chain | ELLIO and ctxsec report the identical full value |
What is not here: no domains, no payload paths, no new user agents, no C2. The scanner UA below is the already-known artefact, not a new one. And note the grading: the IP is an opportunistic multi-purpose host — blocking it costs little but catches one actor, not a campaign; the hash is the more durable artefact, because a retargeted body will change bytes but a replay of this one will not.
ELLIO reported a reconnaissance scanner, and it is the only concrete artefact in the whole case:
| Attribute | Observed |
|---|---|
| User agent | Mozilla/5.0 (compatible; node2shell-check/1.0) |
| Origin | AS44620, registered in Turkey |
| Behaviour | GET /, then 20 POST bodies with Content-Encoding: br in two bursts about two minutes apart: 5 malformed 1,032-byte stubs padded with “A”, and 15 valid Brotli streams across four sizes (largest 65,743 bytes) |
| Key negative | None of the bodies carried the block-switch construct. It is mapping which hosts have a Brotli decode path, not exploiting them |
| Verdict logic | Bases its result on Server / X-Powered-By headers, and does not treat Brotli response compression as proof of vulnerability |
The same source dates its first sighting to 13 June 2026 — four months before the vulnerability had a name, and before the user agent string node2shell-check/1.0 could plausibly exist. These two facts cannot both be true as stated. The likely explanation is that the first-seen date belongs to the ASN or the sensor, not to that user agent. It is a single source, unverified, and recorded here as a discrepancy rather than repeated as a pre-disclosure signal.
| Source | Result | Reading |
|---|---|---|
| ThreatFox | no_result | Tested negative, 26 Oct — the query executed and returned nothing. Re-tested 28 Oct, still no_result for the CVE ID and for both product names |
| URLhaus | no_results | Tested negative |
| OTX | 2 pulses | Both generic CVE coverage (“Brotli Decoder RCE Affecting Node.js, Bun and Python”; a “Known_Cve” list). No actor, no infrastructure |
| CISA KEV | PRESENT — added 27 Oct | Catalog 2026.10.27, 1,760 entries; dueDate 2026-11-10. This row read “absent” on 26 Oct and is the single largest correction in the update above |
| EPSS | 0.55483 (99.03rd pct, 27 Oct) | Up from 0.3278 on 25 Oct. The model re-priced on the KEV listing; it does not add technical information about reachability |
These feeds index malicious infrastructure and blocklists. They do not index attacking hosts seen by canaries, so a zero here was always a statement about four feeds, not a survey — which is exactly how the 26 October indicator set was missed. The lesson generalises: vendor and honeypot telemetry is where attacking hosts appear first; the feeds are where they arrive later, if at all.
Collected 26 October. Reachability stated per source, because a blocked page and an empty page both return 200.
| Source | State | Contribution |
|---|---|---|
Mastodon — #node2shell tag timeline | read 30 posts | The richest channel. Practitioner scepticism, the HTTP2Shell rename, tooling launches, and organisations disclosing service suspensions |
| Hacker News (Algolia) | read | Submissions from 23 Oct; modest engagement compared with PHP2Shell — the topic has not captured general attention |
| Vendor and research blogs | read | Node.js, Imperva, ELLIO, Sysdig, Phoenix Security, Okta, Vercel, Tanium, Strobes, Raven, Aikido |
| The Hacker News (thehackernews.com) | 403 | Returned HTTP 403 from this network — unexecuted, not absent |
| X / Twitter | not usable | Login-walled; only reachable through search indexing. “Not seen on X” is a statement about an index |
| Bluesky | not usable | Direct API 403; posts visible only via bsky.brid.gy bridges that surface on Mastodon |
| not usable | 403 interstitial; not read | |
| Telegram | unexecuted | No channel identified for this topic; not queried |
| Dark web / closed forums | unavailable | No Tor access in this environment. Substituted with abuse.ch, OTX and public reporting. Any claim about closed-forum interest in this CVE would be fabricated, so none is made |
Several posts circulate a “9 detections, 8 IOCs” claim from a commercial intel page. The page itself advertises 31 indicators and 9 rules, all paywalled. The number in circulation does not match the source, and none of the values are visible. Do not carry counts you cannot inspect.
2026.10.27, 1,760) and the case’s earlier “absent” reading corrected as a result.CVE-2026-116579 (Autobahn-Python, unbounded Brotli decompression, CVSS 5.9) is a different bug in a different product. It appears in any “brotli” search and must not be merged into this case.node2shell-check scanner observation.2026.10.23 (fetched directly); FIRST EPSS API, 25 Oct value; ThreatFox, URLhaus and OTX queried 26 Oct.2026.10.27, fetched 28 Oct: 1,760 entries, CVE-2026-111497 added 27 Oct, due 10 Nov. Announcement: 27 Oct alert. Primary.exploitation: active timestamped 27 Oct 19:53 UTC. Primary.94.154.43.7: noise: true, malicious, last seen 27 Oct. Queried directly.ALAS2023-2026-3334 (important): brotli and ALAS2023-2026-3330 (nodejs24), 24 Oct. Primary. AL2 carries no matching item — not established, not cleared.AC:High, RHSA-2026:90501 / 90495 / 90449 on 26 Oct, “no public, reliable RCE PoC”. Primary.11.2.10-h16 fixes PHP only. Primary.--allow-fs-read over-grant fix carries no CVE in the release notes. Primary.--allow-fs-read / --allow-fs-write over-grant. Confirms it is a distinct October item, not the November CVE-2026-123873 gap. Primary.