Upwind Security MDR

🚨High - libkcapi One-shot AIO Use-After-Free Memory Corruption (CVE-2026-71226) libkcapi’s one-shot async I/O (AIO) path can return an error before all submitted AIO control blocks are drained. This leaves in-flight ops that later trigger kernel writes into caller-provided output buffers after they’ve been freed/reused, causing UAF-driven memory corruption and potential code execution. Non-AIO paths aren’t affected. 👉Affected: libkcapi / libkcapi-main; openshift/ose-rhel-coreos-8, openshift/ose-rhel-coreos-9

🚨Critical - Beta9 Tenant Escape via Volume ID Bind-Mount Traversal (CVE-2026-120308) Beta9 gateway builds a hostPath bind mount from the user-supplied volume identifier when deploying a function/endpoint/task queue, without validating ownership or ../ traversal. A crafted volume ID mounts the worker filesystem root into the sandbox with write access, enabling cross-tenant volume access and worker file tampering. 👉Affected: github.com/beam-cloud/beta9 (Beta9) <= 2026-11-01

🚨Critical - Langchain-Chatchat Unauth Tool Invocation RCE (CVE-2026-75128) Langchain-Chatchat exposes POST /tools/call to invoke registered tools by name without auth. Since the shell tool is always registered, a single unauthenticated request can execute arbitrary OS commands with server privileges (often root in the shipped Docker image). 👉Affected: langchain-chatchat <= 0.3.1.3

🚨High - util-linux mount(8) TOCTOU in restricted bind mounts (CVE-2026-78410) util-linux mount(8) has a TOCTOU flaw where the bind mount source path read from /etc/fstab can be swapped before the privileged mount occurs. A local user who can replace the authorized source (or a writable ancestor) can redirect the SUID mount to bind an attacker-chosen directory; with X-mount.owner/X-mount.group/X-mount.mode set, root may chmod/chown the redirected inode. 👉Affected: util-linux (restricted bind mounts via /etc/fstab)

🚨High - util-linux nsenter FD leak enables root cgroup migration (CVE-2026-78408) util-linux nsenter with --join-cgroup opens target cgroup.procs as root but doesn’t close the FD across later setns/credential changes and execve(). Inherited attacker-controlled code keeps root creds on that FD, allowing cgroup moves of host processes and killing unrelated root services. Runs without --join-cgroup aren’t affected. 👉Affected: util-linux (nsenter --join-cgroup)

🚨HIGH - Go mime.WordDecoder DecodeHeader Quadratic CPU DoS (CVE-2026-42504) Go stdlib mime.WordDecoder.DecodeHeader can hit O(n^2) behavior when parsing MIME headers packed with many invalid encoded-words. Remote attackers can send crafted email/HTTP headers to drive excessive CPU usage and trigger denial of service. 👉Affected: Go standard library (mime) in golang/go-toolset packages (versions with vulnerable DecodeHeader)

🚨High - glibc gettext LANGUAGE Directory Traversal + .mo Crash/UB (CVE-2026-84243) In GNU C Library gettext, LANGUAGE entries containing '/' or equal to '..' (when not in AT_SECURE) allow directory traversal to load message catalogs from attacker-chosen paths. A crafted .mo can crash via out-of-range offsets or trigger undefined behavior via mismatched printf format directives, potentially leading to code execution with the program’s privileges. AT_SECURE contexts are not affected. 👉Affected: glibc >= 2.0

🚨High - Ivanti Connect Secure dsinvoked DSCache Trust Privilege Escalation (CVE-2026-111357) Ivanti Connect Secure 22.8R2.4 mishandles trust of DSCache configuration data consumed by the dsinvoked broker. An attacker with limited access can tamper DSCache to influence dsinvoked execution and escalate privileges to root; reportedly exploited in the wild (Oct 2026). 👉Affected: Ivanti Connect Secure 22.8R2.4

🚨 HIGH - Sandbox Escape & Host RCE in ServiceNow AI Platform (CVE-2026-103093) An improper access control flaw (CWE-284) in the ServiceNow AI Platform allows authenticated administrative users to bypass Rhino sandbox restrictions. Attackers can escape the execution environment and execute arbitrary code on the underlying host operating system, leading to host-level compromise. 👉 Affected: ServiceNow AI Platform < Yokohama Patch 13 Hot Fix 4a, Zurich Patch 10 Hot Fix 3b / Patch 12 / Patch 13, Australia Patch 5 | Hosted instances updated automatically; self-hosted/partner customers should apply patches immediately

🚨Critical - Canonical LXD HTTPS Listener Cert Chain Auth Bypass (CVE-2026-112816) LXD’s HTTPS listener improperly validates client certificate chains, allowing an unauthenticated attacker to present a crafted chain that is accepted as a trusted client. This enables full administrative access to the LXD cluster and all managed VMs/containers. 👉Affected: Canonical LXD < 4.0.15, 5.0.x < 5.0.11, 5.21.x < 5.21.9, 6.x < 6.10.1 | Upgrade to 4.0.15 / 5.0.11 / 5.21.9 / 6.10.1

🚨Critical - Apache MINA JavaNativeMessageDecoder Untrusted Deserialization RCE (CVE-2026-118636) Apache MINA 3.0 org.apache.mina:mina-core JavaNativeMessageDecoder deserializes attacker-controlled frame bytes via ObjectInputStream.readObject() with no class filter. A malicious peer can send a crafted serialized payload to trigger gadget-chain remote code execution. MINA 2.2.x is not impacted. 👉Affected: org.apache.mina:mina-core 3.0.0

🚨Critical - AJ-Report Unauth RCE via Forged Share-Token + GroovyClassLoader (CVE-2026-120209) AJ-Report uses a hard-coded JWT secret to sign Share-Tokens, letting attackers forge valid tokens offline. A minted token with reportCode='dataSet' can hit the testTransform endpoint, which feeds attacker-controlled scripts into GroovyClassLoader for unauthenticated RCE. 👉Affected: AJ-Report <= 1.7.1

🚨Critical - Jpom Super-Admin Auth Bypass via JPOM-USER-TOKEN Condition Injection (CVE-2026-120201) Jpom server org.dromara.jpom:server auth trusts JPOM-USER-TOKEN and passes it into Hutool Condition parsing; injecting operators like "is not null" bypasses checks and grants super-administrator access. Attackers can then reach Jpom script features for unauthorized command execution. 👉Affected: org.dromara.jpom:server 2.10.0-2.11.12

🚨HIGH - Django PostGIS RasterField Band Index SQL Injection (CVE-2026-1207) Django's contrib.gis raster lookups on RasterField when backed by PostGIS fail to safely handle the band index parameter, allowing attacker-controlled band values to be injected into generated SQL. Exploitation via crafted ORM filter/lookup input can lead to data disclosure, modification, or DoS. Non-PostGIS backends without raster lookups aren’t affected. 👉Affected: Django 6.0 < 6.0.2, 5.2 < 5.2.11, 4.2 < 4.2.28 | Upgrade to 6.0.2 / 5.2.11 / 4.2.28

🚨Critical - Google Chrome Views Use-After-Free Sandbox Escape RCE (CVE-2026-79047) Google Chrome’s Views component has a use-after-free reachable via a crafted HTML page; with user interaction (social engineering) an attacker can trigger UAF and gain code execution outside the sandbox. Not exploitable without the victim opening attacker-controlled content. 👉Affected: Google Chrome < 152.0.7977.65 | Upgrade to 152.0.7977.65

🚨High - node-forge ASN.1 BIT STRING Deep Nesting DoS (CVE-2026-119191) Forge (node-forge) asn1.fromDer() may decode ASN.1 embedded inside BIT STRING values without consistently enforcing the configured maxDepth. Crafted DER with deeply nested BIT STRINGs triggers excessive recursion, high CPU, and quadratic memory growth, causing DoS. 👉Affected: node-forge <= 1.4.0 | Upgrade to 1.4.1

🚨Critical - pgAdmin 4 Import/Export \copy TO PROGRAM RCE (CVE-2026-116241) pgAdmin 4 Import/Export Data wraps user queries in psql \copy (...) TO and tries to enforce balanced parentheses. With standard_conforming_strings=off, backslashes in double-quoted identifiers can mask a closing “)”, letting an attacker inject TO PROGRAM so psql runs arbitrary OS commands as the pgAdmin service account. Default standard_conforming_strings=on blocks this path. 👉Affected: pgAdmin 4 < 9.19 | Upgrade to 9.19

🚨 Actively Exploited Access Bypass Flaw in Drupal Commerce Core (CVE-2026-118539) An access bypass / missing authorization flaw in Drupal Commerce Core allows unauthenticated attackers to manipulate orders, tamper with prices, bypass payment verification, or complete orders without full payment. Drupal rates this Highly Critical (21/25) and confirms active in-the-wild exploitation. 👉 Affected: Commerce Core 8.x-2.x < 8.x-2.41 and 3.x < 3.3.13 | Upgrade to v8.x-2.41 or v3.3.13 immediately and audit recent order records for price or payment tampering.

🚨Critical - ClickHouse Server Unauth RCE via Integer Overflow Command Injection (CVE-2026-116366) ClickHouse Server has an integer overflow/wraparound in request parsing/validation that can be triggered remotely to corrupt bounds and reach a command injection sink, resulting in unauthenticated remote code execution under the clickhouse-server user. Not exploitable if the vulnerable endpoint is not exposed to untrusted networks. 👉Affected: ClickHouse 26.3 < 26.3.62.3-lts, 26.8 < 26.8.36.3-lts, 26.9 < 26.9.27.3-stable, 26.10 < 26.10.3.7-stable — upgrade to 26.3.62.3-lts / 26.8.36.3-lts / 26.9.27.3-stable / 26.10.3.7-stable (or later) to mitigate