CODERED VTA

Red Hat Issues Important Thrift Security Update for Enterprise Linux AI Platforms

High
Blockchain visualisation
Photo by Shubham Dhage on Unsplash

Red Hat published security advisory RHSA-2026:94503 on 3 November 2026, rated Important, covering the Apache Thrift packages shipped inside Red Hat Enterprise Linux AI 3.4. The advisory addresses two separate flaws, CVE-2026-43871 and CVE-2026-55969, both of which Red Hat describes as having Important impact after being flagged by internal security scans. The corrected build is thrift-0.15.0-8.el9ai, and it is distributed for all four supported architectures of RHEL AI 3.4: x86_64, s390x, ppc64le and aarch64. The update touches the full package family, including thrift, thrift-devel, thrift-glib, thrift-qt, perl-thrift and python3-thrift, so any system carrying the language bindings is in scope and not just the core library.

The vendor has not published exploitation details for either CVE in this errata. There is no attack vector description, no CVSS vector string and no proof-of-concept code referenced in the advisory text, which instead documents the fixed package versions and their SHA-256 checksums. What is known is that the vulnerable code shipped across every architecture build of the platform and that the remediation is delivered purely as a rebuilt RPM set rather than a configuration change or workaround. Red Hat tracks the work internally under RHAI-18641 and notes that Red Hat Insights patch analysis can identify which systems the advisory applies to. The vendor also states plainly that these RPMs are internal build artifacts and are supported only as part of the Red Hat AI application platform, not as standalone components.

Thrift is a cross-language remote procedure call and serialisation framework, so it typically sits in the service-to-service communication layer underneath an AI platform rather than at its edge. That placement means real-world exposure depends on which platform components link against the library and whether those services accept data from untrusted sources. FIRST EPSS places both CVE-2026-43871 and CVE-2026-55969 at a 1.0 percent probability of exploitation within the next 30 days, and no exploitation status beyond that probability is available for either flaw, which is not the same as evidence that they are not being used. The practical consequence for operators of RHEL AI 3.4 is a known-vulnerable dependency, which Red Hat rates as Important, sitting in the runtime of an AI workload platform until the corrected packages are in place.

Attack Surface

Server OS, Supply Chain (Third-party vendors), Infrastructure

Tactics

Initial Access, Impact

Techniques

  • T1190 – Exploit Public-Facing Application
  • T1195.001 – Compromise Software Dependencies and Development Tools
  • T1499 – Endpoint Denial of Service

SuperPRO's Threat Countermeasures Procedures

  1. Update all Red Hat Enterprise Linux AI 3.4 hosts to thrift-0.15.0-8.el9ai, the fixed build published in RHSA-2026:94503 for x86_64, s390x, ppc64le and aarch64, which remediates CVE-2026-43871 and CVE-2026-55969.
  2. Update the full package family, not just the core library: thrift, thrift-devel, thrift-glib, thrift-qt, perl-thrift and python3-thrift all ship at 0.15.0-8.el9ai, and leaving the Python or Perl bindings at an older release leaves vulnerable code on disk.
  3. Inventory affected systems with 'rpm -qa | grep -i thrift' on each RHEL AI 3.4 node and compare the output against 0.15.0-8.el9ai, or use the Red Hat Insights patch analysis view referenced in RHSA-2026:94503 to list systems the errata applies to.
  4. Verify package integrity before deployment against the SHA-256 values in the errata, for example 6e34a72e131d1efc5f2b746e6f703e9f5b1b8a7bab2abbd1dd3736db35ebe309 for thrift-0.15.0-8.el9ai.src.rpm and 3a3143593eb713cf2500d9b7fc14b92b46680dc04799ee3a4d1f50ca66c5f8db for the x86_64 thrift binary RPM.
  5. Confirm all previously released RHEL AI 3.4 errata are applied first, as Red Hat states this update assumes prior errata are in place, and follow the package update procedure in https://access.redhat.com/articles/11258.
  6. Restrict network reachability of Thrift RPC listeners used by platform services, typically TCP/9090 by default, to internal management ranges or the cluster network so the library is not reachable from untrusted segments while the update is being staged.
  7. Treat the thrift RPMs as components of the Red Hat AI application platform only, per the advisory, and remove or avoid any standalone reuse of these builds in non-RHEL AI workloads where they will not receive platform errata.

Source

Code Red Cyber / VTA – coderedcyber.ai