CVE-2026-45736

Impact

The websocket.close() implementation is vulnerable to uninitialized memory disclosure when a TypedArray is passed as the reason argument.

Proof of concept

import { deepStrictEqual } from 'node:assert';
import { WebSocket, WebSocketServer } from 'ws';

const wss = new WebSocketServer(
  { port: 0, skipUTF8Validation: true },
  function () {
    const { port } = wss.address();
    const ws = new WebSocket(`ws://localhost:${port}`, {
      skipUTF8Validation: true
    });

    ws.on('close', function (code, reason) {
      deepStrictEqual(reason, Buffer.alloc(80));
    });
  }
);

wss.on('connection', function (ws) {
  ws.close(1000, new Float32Array(20));
});

Patches

The vulnerability was fixed in [email protected] (https://github.com/websockets/ws/commit/c0327ec15a54d701eb6ccefaa8bef328cfc03086).

Credits

Credit for the private and responsible disclosure of this issue goes to Nikita Skovoroda.

Remarks

Although the calculated CVSS severity is medium, the actual severity is believed to be low, as the flaw is only exploitable through misuse that is unlikely in practice.

Resources

• https://github.com/advisories/GHSA-58qx-3vcg-4xpx
• https://www.cve.org/CVERecord?id=CVE-2026-45736

Root has patched CVE-2026-45736 in the ws package for Root:npm. Multiple fixed versions available.

ws is an open source WebSocket client and server for Node.js. Prior to 8.20.1, the websocket.close() implementation is vulnerable to uninitialized memory disclosure when a TypedArray is passed as the reason argument. This vulnerability is fixed in 8.20.1.

TuxCare rebuilt ws to address 1 CVE. Fixed in 8.20.0-tuxcare.1.

TuxCare rebuilt ws to address 1 CVE. Fixed in 8.18.0-tuxcare.1.

TuxCare rebuilt ws to address 2 CVEs. Fixed in 8.16.0-tuxcare.2.

TuxCare rebuilt ws to address 3 CVEs. Fixed in 8.13.0-tuxcare.1.

TuxCare rebuilt ws to address 1 CVE. Fixed in 8.11.0-tuxcare.2.

TuxCare rebuilt ws to address 1 CVE. Fixed in 8.5.0-tuxcare.2.

TuxCare rebuilt ws to address 1 CVE. Fixed in 8.2.3-tuxcare.2.

TuxCare rebuilt ws to address 2 CVEs. Fixed in 8.18.3-tuxcare.1.

TuxCare rebuilt ws to address 2 CVEs. Fixed in 8.17.1-tuxcare.1.

TuxCare rebuilt ws to address 2 CVEs. Fixed in 8.14.2-tuxcare.1.

TuxCare rebuilt ws to address 2 CVEs. Fixed in 8.18.2-tuxcare.1.

These are all security issues fixed in the kimi-code-0.41.0-1.1 package on the GA media of openSUSE Tumbleweed.

These are all security issues fixed in the zcode-3.14.3-1.1 package on the GA media of openSUSE Tumbleweed.

This update for kimi-code fixes the following issues:

Changes in kimi-code:

Update to version 0.42.0:

• Add an experimental Updates panel with paginated progress messages from the main agent and subagents, enabled with KIMI_CODE_EXPERIMENTAL_NOTIFY_USER=1
• Turn the subagent model pool, Remote Control and the minidb session-index read model with the global search worker always on; their experimental flags and opt-out variables are gone, search persistence is now tuned via the [database] section
• Give tower worker and reviewer briefings the full mission context, follow the subagent timeout for tower agents, and show each background agent's model in /tasks
• Retrieve file search matches beyond the first 100 results, accept HEIC, HEIF and BMP images with Kimi-served models, and warn in kimi -p when project MCP servers are skipped in an untrusted folder
• Read files with configurable character limits and resumable long-line reads, decode malformed UTF-16 with a warning, and preserve image and video filenames in session history
• Rework the dashboard: deletable sessions, reorderable media rail, collapsed tool calls revealed with Ctrl-O, and less jank on long conversations and session reloads

Update to version 0.41.0:

• Add experimental tower mode for multi-agent orchestration and Remote Control for reaching a local web session remotely
• Add the WaitFor tool so the agent can wait for a background task inside the current turn, and an optional fork parameter that starts a subagent from a snapshot of the caller's history
• Support two OAuth login methods, kimi.ai and kimi.com, and activate several skills in one prompt
• Enable the subagent model pool in every launch mode by default and make turn-level file history unconditional
• Add a dangerous-command guard that always asks before shutdown, reboot or rm -rf in manual and YOLO permission modes, disabled with [permission] dangerous_command_guard; auto mode never blocks
• Remove kimi web --allow-remote-terminals; PTY terminal routes are served on loopback binds only
• Require Edit and Write to read a file first, and reject a write when the file changed on disk since that read
• Rework the bundled web dashboard: multi-tab right sidebar, Open/Done/Workspaces session tabs, a plugin marketplace panel and mobile bottom sheets
• ... see upstream's release notes for the full list

• Vendor the runtime dependencies upstream stopped inlining in 0.39.0:
  • Add ws, qrcode and qrcode's own pngjs and dijkstrajs as sources, installed into the package's private node_modules
  • Unpack the npm tarball directly instead of running npm install, which would resolve the new dependencies against the registry and pull in qrcode's CLI-only yargs tree
  • Declare the vendored trees with Provides: bundled(...)
• CVE-2026-48779: ws denial of service from a high volume of tiny WebSocket fragments; the ws inlined in the old bundle predated the 8.21.0 fix, the vendored ws is 8.21.3 (boo#1268927)
• CVE-2026-45736: ws uninitialized memory disclosure via websocket.close() with a TypedArray, fixed by the same ws version (boo#1269951)
• Load the compiled node-pty addon during the build check as well