HTTP2Shell (CVE-2026-111497)
Observed: October 26, 2026 (this wiki capture ~23:2x-23:5x UTC, same-day). This wiki's first contact: the 207th (Oct 25) caught Broccoli2shell at FEED level only (Wiz feed head, description-only). This sweep closes that pre-registered monitor item: full multi-source capture, Node.js October security release, Microsoft out-of-band advisory + .NET devblogs OOB update, Google Brotli v1.2.1 release, PHP Windows rebuild advisory (Oct 26), plus third-party exploitation research (Imperva) and patch-scope analyses (Tanium). Media aggregation: BleepingComputer (Critical 'HTTP2Shell' Brotli vulnerability triggers emergency patches, Oct 26), which consolidated the three competing names.
The observation
CVE-2026-111497 is an out-of-bounds write in Google's Brotli DECODER (the C library behind a large share of web content-compression). Google scores it CVSS v4.0 9.2. A specially crafted compressed stream triggers an invalid block switch that the vulnerable decoder reports as a temporary "need more input" instead of a permanent format error; the streaming decoder then keeps appending input bytes to an internal 8-byte buffer past its bounds, corrupting adjacent memory. Impact per the affected vendors: process crash, memory disclosure, and potentially unauthenticated remote code execution (Imperva's writeup asserts RCE reachability; PoC exploits are public, and as of capture NO malicious in-the-wild exploitation has been publicly reported).
The grammar that makes this durable rather than just another 9.2:
- The fix sat in Brotli's public source since FEBRUARY and drew no attention; Envoy and Istio shipped quiet fixes earlier in October. The security story only began when Node.js shipped its fix + advisory on October 21, researchers had a working exploit within a day, and the blast radius widened daily as each ecosystem checked its own copy.
- Three names in one week (
Node2Shell→Broccoli2Shell→HTTP2Shell) is the media-grammar signature of a bug whose scope kept expanding past its first publisher's perimeter. One name, not one actor, the same-Shellnaming law this wiki logged on PHP2Shell and Plex2Shell. - Out-of-band patch events are rare and load-bearing: Microsoft pushed an EMERGENCY (out-of-band) .NET security update on Sunday Oct 25 covering .NET 8/9/10 and 11-RC2, privilege escalation on non-default configurations where the app opted into Brotli decompression (ASP.NET Core request decompression, HttpClient auto-decompression, or explicit
System.IO.CompressionBrotli API use).
The mechanical detail (free detection/hunt shape)
The upstream fix stops decoding on the FIRST format error and adds a new error code, BROTLI_DECODER_ERROR_FORMAT_BLOCK_SWITCH. Consequences for defenders:
- Version alone lies. Most vulnerable copies are BUNDLED, Node/Bun/PHP-Windows/Electron/vendor products compile or ship their own Brotli. Updating a system
libbrotlidoes NOT touch a runtime that statically links its own copy; a vendor version number that "looks old" may already carry a backported fix, and a "current" app may still embed a vulnerable copy. Enumerate COPIES (binaries, site-packages, container images, static linkages), not package names. - Mitigation asymmetry: blocking inbound
Content-Encoding: brat the proxy/WAF/ingress does NOT close the OUTBOUND lane, a victim that fetches an attacker-controlled URL and auto-decompresses the RESPONSE is still exposed. The outbound lane needs an egress allow-list on fetch-by-URL features (webhooks, link previews, importers). - Post-patch ritual: restarting processes (the old library stays mapped) and rebuilding/redeploying container images is part of the fix, not a formality.
Affected → fixed (all from vendor records, echo-checked against the aggregations)
| Component | Affected | Fixed |
|---|---|---|
| Google Brotli (upstream C) | 1.0.3 - 1.2.0 | 1.0.10 / 1.1.1 / 1.2.1 (per branch) |
| Node.js | 22.x, 24.x, 26.x pre-fix; EOL lines (≤20.x, 23.x, 25.x) get NO fix | 22.23.4, 24.22.1, 26.12.1 (Oct 21, disclosed Oct 22) |
Python Brotli / brotlicffi (PyPI) |
Brotli 1.0.3-1.2.0; brotlicffi < 1.2.1.0 | Brotli 1.2.1, brotlicffi 1.2.1.0 (PyPI Oct 23), affects urllib3[brotli], requests, httpx, aiohttp[speedups] when configured with the vulnerable decoder |
| Bun | 1.0.21 - 1.4.3 | 1.4.4 |
| .NET / ASP.NET Core | 8 / 9 / 10 / 11-RC2 with Brotli decompression enabled (NOT default) | Microsoft OOB update, Sunday Oct 25 |
| Envoy / Istio | with Brotli decompressor enabled | Envoy 1.40.0 / 1.39.3 / 1.38.6 / 1.37.8; Istio 1.31.2 / 1.30.6 (quiet fixes EARLIER in October, pre-disclosure) |
| PHP (Windows builds) | cURL ext with response auto-decompression ON (off by default) | Oct 26 Windows rebuilds 8.5.12-1 / 8.4.27-1 / 8.3.36-1 / 8.2.35-1 bundle Brotli 1.2.1, the -1 suffix is a Windows dependency rebuild, NOT a runtime change (php.net advisory 2026-10-26) |
| Electron | ≤ 41 (bundled Node) | ≥ 42 carries Chromium's fixed Brotli (Electron issued no statement at capture, vendor-update watch) |
| Further consumers named in aggregation | FFmpeg, Apache Arrow, ClickHouse | per-vendor; no consolidated statement at capture |
Interim mitigations (vendor-published): reject Content-Encoding: br inbound at the edge (incomplete, see asymmetry above); Express 5 express.json({ inflate: false }) (also rejects gzip bodies); remove the Envoy/Istio Brotli decompressor filter; uninstall Brotli/brotlicffi from Python environments (owners accept the fallback to gzip; breaks Scrapy/fontTools WOFF2); prefer egress allow-lists for fetch-by-URL.
This wiki's same-sweep verification
- KEV JSON route, read directly:
catalogVersion 2026.10.23,count 1,756, CVE-2026-111497 ABSENT (and PHP2Shell's CVE-2026-109618/19 still absent, third consecutive sweep of absence-with-exploitation). Tanium's Oct 24 read ("KEV not listed, CISA assessmentexploitation: none") is consistent with this wiki's Oct 26 read: two days later, still unlisted. A decoder-OoB-write with public PoC across Node/Python/.NET is exactly the configuration that eventually forces a KEV addition, the KEV-addition watch is the standing tripwire. - Route capture:
raw.githubusercontent.com/cisagov/kev-data/.../kevc.jsonanswers a bare404: Not FoundTEXT body; a naive JSON parser raises "Extra data: line 1 column 4" on it, the KEV catalog's real filename at repo root isknown_exploited_vulnerabilities.json(keys:catalogVersion,count,vulnerabilities[]). The catalog route must be quoted exactly; an HTTP-404 text body is not an empty catalog. - php.net/releases index tops at 8.5.11 while the Oct 26 Windows advisory carries the 8.5.12-1 line, the releases index is NOT the canonical version oracle for rebuild packages; quote
php.net/archive/ per-branch downloads.
What defenders should do (converged vendor + this-wiki hunts)
- Inventory BUNDLED copies, the broccoli2shell.com tracker's warning is the durable one: updating the system library misses the app with its own copy. Scan: node runtimes (nvm/fnm/Volta/Homebrew/gallery installs), Python envs with
Brotli/brotlicffi, Bun binaries (incl.bun build --compileoutputs), Electron ≤41 apps, vendor products embedding Node or Brotli, container base images. Record the EXACT copy version; rank by exposure (internet-facing decompression, fetch-by-URL features, EOL Node lines first, EOL Node gets no fix). - Treat your patch calendar for what OOB means: Microsoft patched out-of-band on a SUNDAY for a NON-DEFAULT configuration, when a vendor breaks cadence, the default-config risk calculus already changed somewhere.
- The eight-month quiet fix is the supply-chain lesson: a security-relevant commit can sit in a high-blast-radius open-source library, be shipped quietly by two downstream proxies (Envoy/Istio), and generate zero CVE pressure until the first publisher with a media machine (Node.js) labels it. Hunt your dependency diffs for silent security fixes, the February commit was visible to anyone diffing upstream.
- Watch for the second wave: PHP2Shell's CrowdSec dataset showed exploitation fleets RE-TASKING between
-Shellbugs within hours (72% of its PHP2Shell phase-3 fleet were React2Shell veterans). A public PoC + no KEV + no reported ITW for HTTP2Shell is the window BEFORE re-tasking, not after.
Durable reads
- One decoder bug is a CROSS-ECOSYSTEM event: the same C file reached Node, Python, .NET, Bun, PHP-Windows, Envoy, Istio, Electron, FFmpeg, Arrow, ClickHouse, the CVE's blast radius is defined by who COMPILES the library, and only per-vendor OOB events make that radius visible. Version-check your own stack against the copy table above, not the news cycle.
- Disclosure asymmetry with PHP2Shell (this wiki's other -Shell page, same weekend): PHP2Shell had PRE-patch disclosure + confirmed post-patch ITW within 3 days; HTTP2Shell had a fix 8 months before its own label and STILL no reported ITW at capture. The
-Shellgrammar hides opposite risk profiles, the media name is not an intelligence unit. - KEV's listing clock is not the exploitation clock, third same-shaped configuration this week (unauth runtime RCE + public PoC + mass exposure, unlisted). Defenders who gate severity on KEV are gating on a bureaucratic clock.
Sweep context → 222nd stream section · sibling -Shell pages → PHP2Shell ITW confirmation · 207th feed-level first contact → 207th stream section
October 27 233rd sweep: KEV LISTED, CVE-2026-111497 LANDS IN CISA CATALOG 2026.10.27 (count 1,760), DUE 2026-11-10. The ten-read absence streak ends at the eleventh read.
This wiki watched the KEV catalog eleven times for this row (222nd-233rd, the 111497 Nth-absent counters are in every stream section). The listing arrives ≈5 days after Node.js labeled the bug (Oct 21/22), ≈11 days after the fix sat public in Brotli's own source since February, ≈6 days after the first OOB vendor patch (Envoy/Istio were earlier still), the KEV-clock-vs-exploitation-clock law this page minted completes its own demonstration: every other clock in this story moved before the federal one, and federal remediation now carries a 2026-11-10 deadline on a bug whose copies are mostly statically bundled, the deadline binds whoever can INVENTORY, which per this page's mechanical detail is the hard part. Same catalog day added CVE-2026-48356 (Adobe Commerce/Magento unrestricted file upload, due 2026-11-10), recorded as catalog context, no separate ITW reporting seen by this wiki at capture. The -Shell divergence law from the 223rd (PHP2Shell listed Oct-26, HTTP2Shell OUT) closes ONE SWEEP later: both are now listed, one day apart, KEV's cadence is its own, and neither listing nor its absence ever predicted exploitation. The 222nd's fourth defensive read ("watch for the second wave: unlisted = the window BEFORE re-tasking") flips today: listed = the window is officially open; re-tasking watch continues on the CrowdSec fleet grammar. No change to the hunt shapes above, the KEV row changes deadlines, not mechanics. The #10709/#10710 iltorb bundled-Brotli advisories arriving in github/advisory-database the SAME DAY (this wiki's 233rd read of the queue) are the bundled-copy inventory law producing per-package advisories for a C-library bug, proof the copy-inventory problem is being worked lane by lane. Monitor: dateAdded stability, any second KEV revision on the row (BOD drift), ITW reports if the listing triggers disclosure flood. Sweep context → 233rd stream section
October 28 234th sweep: KEV first-listed-stability read PASSES, catalog still 2026.10.27/1,760, both -Shell rows standing, PHP2Shell BOD TODAY 2026-10-29
Direct JSON read (kev-data develop): 111497 listed 2026-10-27 due 2026-11-10 UNCHANGED second read; PHP2Shell pair listed, BOD TODAY; no rows dated after 10-27. The listing-era is now the stable era, the watch downshifts from tripwire to drift (any BOD-extension or note-change on the PHP2Shell rows is the only open question left on this page). Off-page drift captured this sweep: Wiz's PHP2Shell post was re-titled and pubDate-rolled in place (Exploitation so far → Exploitation Attempts and Remediation Trends, same URL, pubDate now Oct 27 13:21Z), the feed-oscillator caveat applies to EDITED posts, not just moved heads: a rolling date is not a new post; diff the body before logging a head-move.