CVE-2026-108958
MediumCVSS 6.3Exploitation Probability (EPSS)
Low risk16th percentile - higher than 16% of all known CVEs
Summary
A weakness was identified in TensorLayer up to 2.2.4. The function load_hdf5_graph in tensorlayer/files/utils.py is affected; manipulating the model_config argument leads to deserialization. The attack can be executed remotely, and the exploit has been made public.
Risk Assessment
A remote attacker can exploit deserialization to execute unauthorized code or cause application crashes. The public exploit increases the risk of attacks.
Recommendation
It is recommended to update TensorLayer to the latest version or apply available patches. Also, avoid processing untrusted HDF5 files.
Related vulnerabilities
- CVE-2026-115931Critical
Postiz lets a signed-in user ask the backend to run a function on one of their connected channels, but the backend did not restrict which functions could be requested, allowing calls to internal helpers. One helper reads a file from the server's disk (e.g., environment file with the session signing secret, database address, and service credentials), and another makes the backend process exit with a single request. Self-registration is open by default, so no existing account or action by another user is needed.
- CVE-2026-115893Critical
Wasmtime prior to 36.0.19, 48.0.7, and 49.0.4 has a flaw in the Cranelift compiler on aarch64. A br_table instruction with more than ~256,000 targets can be miscompiled, allowing a jump to arbitrary code and sandbox escape.
- CVE-2026-115889Critical
Wasmtime prior to 36.0.19 and 43.0.0 has a flaw in the Cranelift compiler where SSA value indices can silently overflow when compiling large Wasm functions. This can cause miscompilation, type confusion, and sandbox escape.
- CVE-2026-115812Critical
A relative path traversal during model archive extraction in AWS Multi Model Server 1.1.11 and earlier allows remote actors to execute arbitrary code via a crafted model archive. The project is end of life and no fixed version is available.
- CVE-2026-55486Critical
In Goobi viewer from version 25.08 to before 26.05, the IIIF Auth 2.0 endpoints auth/login and auth/token accept an origin query parameter without validation and embed it into the Content-Security-Policy header and an HTML response body as a postMessage argument. Since the attacker controls both calls, the unescaped value yields reflected XSS.
- CVE-2026-53942Critical
In conda/conda prior to commit 9c0f238, a GitHub Actions Pwn Request vulnerability remains in .github/workflows/update.yml. It is triggered by comments containing @conda-bot render or @conda-bot recreate, allowing an untrusted PR to reach Jinja2 template rendering in a privileged token context.
- CVE-2026-46652Critical
In Goobi viewer from version 24.03 to before 26.05, ExternalFilesDownloader.getFilename() extracts the filename from a server-supplied Content-Disposition header and passes it directly to Path.resolve() without stripping directory components. A malicious HTTP server can supply an absolute path or ../ sequences, writing attacker-controlled content to an arbitrary path, leading to remote code execution.
- CVE-2026-46651Critical
In Goobi viewer from version 24.03 to before 26.05, the DownloadTaskEndpoint WebSocket endpoint (/tasks/download/monitor.socket) in goobi-viewer-core accepts download requests from unauthenticated users. This issue has been patched in version 26.05.
- CVE-2026-53921Critical
In odhcpd (the DHCPv6/DHCPv4/RA server used by OpenWrt) prior to version 25.12, the DHCPv6 IA response handler writes past a fixed 512-byte stack buffer (pdbuf) when a single DHCPv6 REQUEST carries enough IA options to nearly exhaust the reply space. Two independent overflow sites exist in the same call chain, both reachable over the network via a single unauthenticated UDP packet. odhcpd runs as root on OpenWrt, and embedded targets often lack stack canaries and ASLR, making code execution realistic.
- CVE-2026-114357Critical
In DJI CrystalSky ZS600C firmware v0.3.9.7 through v0.6.6.9, the Android Debug Bridge daemon (adbd) lacks authentication for a critical function. The device ships with ro.adb.secure=0 and USB ADB enabled by default; once ADB over TCP is enabled, the same unauthenticated shell (uid 2000) is reachable over the network. This provides a foothold for privilege escalation to root.
Original NVD description (English source)
A weakness has been identified in TensorLayer up to 2.2.4. This affects the function load_hdf5_graph of the file tensorlayer/files/utils.py of the component HDF5 Loading. Executing a manipulation of the argument model_config can lead to deserialization. The attack can be executed remotely. The exploit has been made available to the public and could be used for attacks. The project was informed of the problem early through an issue report but has not responded yet.
Vulnerability data from NVD (NIST) · CISA KEV · EPSS

