Seven are rated high, including an account takeover through OAuth2 auto-linking and fork workflows that receive repository secrets. Every release up to 28.2.0 is affected; there is no backport line.
Wasmtime fixes two critical sandbox escapes in its Cranelift compiler
EditorialBy TrustList Editorial
Both are rated 9.9. One affects every Wasmtime release before version 43 on any architecture; the other has been in the aarch64 back end since 2020. Fixed in 36.0.19, 48.0.7 and 49.0.4.
- Cybersecurity
- Vulnerability Management
- Patch Management
- Application Security
About Wasmtime fixes two critical sandbox escapes in its Cranelift compiler
Wasmtime fixes two critical sandbox escapes in its Cranelift compiler
19 October 2026: The Bytecode Alliance has fixed two critical flaws in Wasmtime, the WebAssembly runtime used to run untrusted code in serverless platforms, plug-in systems and edge services, that let a crafted WebAssembly module break out of its sandbox. Both are rated 9.9 under CVSS 3.1 and both sit in Cranelift, the compiler Wasmtime uses to turn WebAssembly into machine code. Versions 36.0.19, 48.0.7 and 49.0.4 carry the fixes.
Not yet independently verified. Both advisories are the Bytecode Alliance's own (CVE-2026-115889 and CVE-2026-115893, published 19 October 2026) and we found no independent analysis. Neither gives a proof of concept or reports exploitation. We will update this when it can be confirmed, and remove this note.
The project published the two advisories on 19 October. Neither reports exploitation.
Update, 11 November 2026: The two advisories now carry CVE numbers: GHSA-897f-g5gm-fmxm is CVE-2026-115889 (CVSS 9.9, fixed in 36.0.19, version 43 not affected) and GHSA-mghf-wjfp-p5xm is CVE-2026-115893 (CVSS 9.9, the aarch64 br_table flaw, fixed in 36.0.19, 48.0.7 and 49.0.4). Neither states a workaround.
How the two flaws work
The first, CVE-2026-115889 (GHSA-897f-g5gm-fmxm), affects Wasmtime before version 43 on every architecture with default settings. Cranelift packed the index of each value into a limited number of bits, and in release builds a very large or deliberately crafted function could overflow that field without any error. The compiler then confused one value with another, which the advisory classes as type confusion leading to a sandbox escape. Cranelift moved to full 32-bit indexes with checks in version 43.
The second (GHSA-mghf-wjfp-p5xm) is specific to 64-bit Arm. A module with more than about 256,000 targets in a single br_table instruction made the aarch64 back end compute the wrong branch target, so the program could jump to code at an offset the attacker chose. The advisory says every Wasmtime version with aarch64 support is affected, going back to 2020.
Which versions to install
- 49.x: update to 49.0.4
- 48.x: update to 48.0.7
- 36.x (long-term support): update to 36.0.19
- 50.0.0: released after the advisories, also fixed
Releases older than these lines get no fix, so anyone on another branch needs to move to one of them.
Who should move first
The exposure is greatest where Wasmtime runs modules that someone else wrote: function-as-a-service platforms, CDN and edge runtimes, plug-in hosts and blockchain-style execution engines. Services that only run their own compiled modules are at lower risk, but the overflow can be triggered by size alone, so a large legitimate module is not proof of safety. Where an update has to wait, the aarch64 flaw can be avoided by running on x86-64 hosts, while the index overflow has no configuration workaround in the advisory. The same advisory list carries lower-rated fixes from 14 and 19 October, including a moderate filesystem sandbox escape through symlinks with a trailing slash, which the same releases address.
Sources
- Bytecode Alliance: Wasmtime advisory GHSA-897f-g5gm-fmxm (CVE-2026-115889), Cranelift value index overflow, 19 October 2026
- Bytecode Alliance: Wasmtime advisory GHSA-mghf-wjfp-p5xm, aarch64 br_table miscompilation, 19 October 2026
- Wasmtime releases on GitHub (read 27 October 2026), 27 October 2026
- Bytecode Alliance: Wasmtime advisory GHSA-897f-g5gm-fmxm (CVE-2026-115889), 19 October 2026
- Bytecode Alliance: Wasmtime advisory GHSA-mghf-wjfp-p5xm (CVE-2026-115893), 19 October 2026
Categories & features
- Cybersecurity
- Vulnerability Management
- Patch Management
- Application Security
TrustList Weekly
The week in software and IT, in one email
The news that matters to buyers, new rankings and our own research. Every Thursday, free, and easy to leave.
You may be interested in
Besides the bundled OpenJPEG overflow, the Python imaging library published 39 more advisories fixed in 13.0.0, including code execution from one crafted FlashPix file. One Arrow memory leak is still unfixed.
CVE-2026-111497 is rated 9.2 by Google and hits every Brotli release from 1.0.3 to 1.2.0. Envoy, the PyPI packages and Debian have shipped fixes; Ubuntu still ranks it medium.
Releases 26.12.1, 24.22.1 and 22.23.4 close ten CVEs, four rated High, and also drop trust in the compromised eMudhra intermediates through the bundled CA store.
PHP 8.5.12, 8.4.27, 8.3.36 and 8.2.35 close CVE-2026-109618 and CVE-2026-109619. Every PHP install serving web requests is at risk, mod_php most of all, and PHP 8.1 and older get no official fix.
Two flaws rated Important hit all three supported lines, a third affects only Tomcat 11, and the fixed releases have been out since 12 October.
Version 155.0.8059.84/.85 patches a critical use-after-free in Prerender and six high-severity bugs, four of them in the V8 engine. Hong Kong's CERT warns of remote code execution.
SA-CORE-2026-014 scores 19 out of 25. Sites on 11.2 or 10.5 and older have no patch; uninstalling the Field UI module blocks the flaw until an upgrade is possible.
More on TrustList
Everything here links back to the same verified catalogue. Pick your next stop.
- More CybersecurityThe ranking for this subject
- CompaniesAgencies, consultancies and IT service providers, ranked by verified reviews.
- ProductsSoftware and SaaS with pricing, features, integrations and alternatives.
- AwardsAnnual recognition decided by verified reviews and an independent jury.
- LaunchesNew products and releases, voted up by the community every day.
- AI ModelsBenchmark scores and community ratings for every major model.
- RequestsBuyers describe what they need; vendors respond directly.
- PeopleReviewers, authors and makers with public profiles.
- ComparePut up to four listings side by side before you shortlist.