CVE-2026-44509
Rsync is a file-copying tool that uses a delta-transfer algorithm to synchronize remote and local files. In versions prior to 3.4.3, previou
CVSS
6.3
Medio
EPSS
0.1%
p0
KEV
—
Exploit Today
0
0-100
Publicado: 20 jul 2026 · Última mod.: 21 jul 2026 · CWE-59 · CWE-367 · CWE-863
Sin historial EPSS suficiente todavía.
Rsync is a file-copying tool that uses a delta-transfer algorithm to synchronize remote and local files. In versions prior to 3.4.3, previous bug fixes for symlink races in open() calls missed races in other path based system calls like chmod() and chown(). For rsync daemons with "use chroot = no" this allows an attacker with local filesystem access to change permissions, ownership or timestamp on a file outside the exported module. This issue is fixed in version 3.4.3.
CVECVSSEPSSKEVRExplotTítuloVis.
CVE-2026-471216.1 MED—
——0Sparkle is a software update framework for macOS. Prior to version 2.9.2, `Autoupdate/SUBinaryDeltaApply.m` enforces `relativePath.pathComponents containsObject:@".."` and rejects writes whose immediate parent directory IS itself a symbolic link, but does not detect symlinks deeper in the relative path. `Autoupdate/SPUSparkleDeltaArchive.m`'s `extractItem:` will create symlinks in the destination tree from archive content (no `..` check on the symlink target), and a subsequent `Extract` item targeting `<symlink>/foo/bar` then escapes the destination tree via `fopen(path, "wb")` because the kernel resolves the intermediate symlink during the open call. This is a defense-in-depth issue: exploitation requires a maliciously-crafted `.delta` that passes EdDSA signature verification, i.e. EdDSA private-key compromise. With the AppInstaller running as root for system-domain installs, it gives the holder of a stolen signing key arbitrary file write at root level via the delta-apply path, which is a strictly broader primitive than the "drop-in replacement bundle" install they would otherwise have. Version 2.9.2 contains a patch for the issue.3hCVE-2026-471286.1 MED0.5%
——0nono is software that allows users to run AI agents in a zero-latency sandbox. Prior to version 0.55.0, the nono Landlock/seccomp policies allow access to local Unix domain sockets (concrete and abstract). This allows an easy sandbox escape by talking to the per-user systemd dbus socket. Version 0.55.0 patches the issue.4hCVE-2026-64651—1.7%
——1The `@ai-sdk/harness-opencode` tool connects HarnessAgent to OpenCode through a sandboxed bridge. Prior to version 1.0.28, the tool relay authorizes requests from any process whose command line contains an allowed helper script path (`host-tool-mcp.mjs`). This allows untrusted code executing in the sandbox to invoke arbitrary host-exposed tools including secret lookups, deployment operations, and cloud API calls without a corresponding model-authorized tool-call event. Exploitation requires a Linux environment (the vulnerable fallback checks `process.platform === 'linux'` and reads `/proc`); an active harness session with one or more host-provided tools; and untrusted code executing in the sandbox (e.g. a malicious dependency, build script, or lifecycle hook) The fix in version 1.0.28 removes the process-path authorization fallback entirely. Relay requests are now only accepted after exact, short-lived, one-time authorization matching the tool name and input from a bridge-observed model event. Some workarounds are available. Do not run the OpenCode harness on untrusted repositories or with untrusted dependencies, and/or limit host-exposed tools to non-sensitive operations when working with untrusted code.4hCVE-2026-64650—1.7%
——1The `@ai-sdk/harness-opencode` tool is an HarnessV1 adapter backed by @openai/codex-sdk, which drives the codex command line interface. Prior to version 1.0.29, the tool relay authorizes requests from any process whose command line contains an allowed helper script path (the Codex CLI shim). This allows untrusted code executing in the sandbox to invoke arbitrary host-exposed tools, including secret lookups, deployment operations, and cloud API calls without a corresponding model-authorized tool-call event. Exploitation requires a Linux environment (the vulnerable fallback checks `process.platform === 'linux'` and reads `/proc`); an active harness session with one or more host-provided tools; and untrusted code executing in the sandbox (e.g. a malicious dependency, build script, or lifecycle hook) The fix in version 1.0.29 removes the process-path authorization fallback entirely. Relay requests are now only accepted after exact, short-lived, one-time authorization matching the tool name and input from a bridge-observed model event. Some workarounds are available. Do not run the Codex harness on untrusted repositories or with untrusted dependencies, and/or limit host-exposed tools to non-sensitive operations when working with untrusted code.19hCVE-2026-445074.8 MED4.3%
——1Rsync is a file-copying tool that uses a delta-transfer algorithm to synchronize remote and local files. In versions prior to 3.4.3, when using a daemon configured with a chroot, the reverse DNS lookup of the connecting client was performed after the chroot had been entered. If the chroot did not contain the files glibc needs for resolution (/etc/resolv.conf, /etc/nsswitch.conf,
/etc/hosts, NSS service modules), the lookup failed and the connecting hostname was set to "UNKNOWN". Hostname-based deny rules ("hosts deny = *.evil.example") therefore could not match, and an attacker who controlled the PTR record for their source IP could connect from a hostname that the administrator had intended to deny. IP-based ACLs ("hosts allow = 10.0.0.0/8") are unaffected. This issue is fixed in version 3.4.3.4hCVE-2026-442319.1 CRÍ33.4%
——10RT is an open source, enterprise-grade issue and ticket tracking system. Versions prior to 5.0.10, 6.0.0 and above, prior to 6.0.3 contain an information disclosure and privilege escalation vulnerability in the REST 2.0 API. A privileged (non-administrative) user can obtain authentication credentials belonging to other users — including users with administrative privileges — and use those credentials to read data as those users via RT's feed endpoints. The same request that exposes the credentials also rotates them, invalidating previously-distributed feed URLs across the instance. This issue has been fixed in versions 5.0.10 and 6.0.3.4h