RHSA-2026:82061CriticalCVSS 9.1

Red Hat Security Advisory: Red Hat AI Inference Tech Preview 3.5.2 (Gaudi)

Published
October 13, 2026
Last Modified
October 14, 2026

🔗 CVE IDs covered (71)

CVE-2026-73076 →CVE-2026-11940 →CVE-2026-16118 →CVE-2026-54369 →CVE-2026-73072 →CVE-2026-73077 →CVE-2026-73555 →CVE-2026-58011 →CVE-2026-101917 →CVE-2026-11824 →CVE-2026-58010 →CVE-2026-64830 →CVE-2026-71486 →CVE-2026-73066 →CVE-2026-73560 →CVE-2026-55693 →CVE-2026-97689 →CVE-2026-102266 →CVE-2026-102272 →CVE-2026-47167 →CVE-2026-54411 →CVE-2026-66036 →CVE-2026-97687 →CVE-2026-97688 →CVE-2026-14164 →CVE-2026-58015 →CVE-2026-69147 →CVE-2026-84379 →CVE-2026-84382 →CVE-2026-101918 →CVE-2026-102274 →CVE-2026-102598 →CVE-2026-84378 →CVE-2026-103001 →CVE-2026-15588 →CVE-2026-52858 →CVE-2026-58012 →CVE-2026-60002 →CVE-2026-93592 →CVE-2026-102265 →CVE-2026-11822 →CVE-2026-59858 →CVE-2026-64834 →CVE-2026-102269 →CVE-2026-102270 →CVE-2026-73078 →CVE-2026-102275 →CVE-2025-6170 →CVE-2026-73558 →CVE-2026-102267 →CVE-2026-59856 →CVE-2026-41523 →CVE-2026-44432 →CVE-2026-73557 →CVE-2026-44431 →CVE-2026-57456 →CVE-2026-102268 →CVE-2026-102271 →CVE-2026-104874 →CVE-2026-57455 →CVE-2026-73559 →CVE-2026-84381 →CVE-2026-13346 →CVE-2026-58013 →CVE-2026-58014 →CVE-2026-73556 →CVE-2026-84380 →CVE-2026-102273 →CVE-2026-41989 →CVE-2026-47162 →CVE-2026-66039 →

📋 Description

CVE-2025-6170 — libxml2: Stack Buffer Overflow in xmllint Interactive Shell Command Handling CVE-2026-11822 — sqlite: SQLite: Arbitrary code execution via crafted FTS5 full-text search data CVE-2026-11824 — sqlite: SQLite: Arbitrary code execution and crash via heap-based buffer overflow in FTS5 CVE-2026-11940 — python: cpython: CPython: tarfile extraction filter bypass allows escaping the destination directory CVE-2026-13346 — pip: pip: Arbitrary file installation via malicious package indexes CVE-2026-14164 — libarchive: Double-Free Vulnerability in RAR5 Decompression Logic via dangling filtered_buf pointer in init_unpack() CVE-2026-15588 — GDBusServer: glib2: GDBusServer pre-authentication DoS via unbounded SASL line buffering CVE-2026-16118 — xdgmime: heap-based buffer overflow in _xdg_mime_magic_parse_magic_line() in xdgmimemagic.c CVE-2026-41523 — vllm: vLLM: Arbitrary code execution via malicious HuggingFace model CVE-2026-41989 — Libgcrypt: Libgcrypt: Denial of Service and buffer overflow via crafted ECDH ciphertext CVE-2026-44431 — urllib3: urllib3: Information disclosure via cross-origin redirects forwarding sensitive headers CVE-2026-44432 — urllib3: urllib3: Denial of Service due to excessive HTTP response decompression CVE-2026-47162 — vim: Vim: Arbitrary Code Execution via crafted directory names CVE-2026-47167 — vim: Vim: Arbitrary code execution via crafted step-definition patterns CVE-2026-52858 — vim: Vim: Arbitrary code execution via Python omni-completion CVE-2026-54369 — acl: Symlink traversal privilege escalation via libacl functions CVE-2026-54411 — linux-pam: Plaintext password recovery via timing discrepancy in pam_userdb module CVE-2026-55693 — vim: Vim: Out-of-bounds Write in Spell File Word Count CVE-2026-57455 — vim: Vim: Denial of Service via stack out-of-bounds write in spell_soundfold_sofo() CVE-2026-57456 — vim: Vim: Arbitrary code execution via malicious docstrings in Python omni-completion CVE-2026-58010 — glib: buffer over-read in glib/gvariant-serialiser.c via gvs_tuple_is_normal() CVE-2026-58011 — glib: out-of-bounds read in glib/gdatetime.c:g_date_time_get_ymd via invalid GDateTime CVE-2026-58012 — glib: buffer over-read in g_regex_replace() via glib/gregex.c:string_append() and g_utf8_next_char() CVE-2026-58013 — glib: buffer over-read in glib/giochannel.c via "g_io_channel_read_line_backend" CVE-2026-58014 — glib: off-by-one error in glib/gkeyfile.c via "g_key_file_get_locale_string_list" CVE-2026-58015 — glib: path traversal in glib/gio/gdbusauthmechanismsha1.c via keyring_lookup_entry and mechanism_client_data_receive CVE-2026-59856 — vim: Vim: Arbitrary code execution via crafted PHP file in omni-completion CVE-2026-59858 — vim: Vim: Arbitrary command execution via crafted tags file in C omni-completion CVE-2026-60002 — openssh: OpenSSH: Use-after-free vulnerability during host key re-exchange on the client side CVE-2026-64830 — ffmpeg: FFmpeg: Arbitrary code execution via heap buffer overflow in VobSub subtitle demuxer. CVE-2026-64834 — FFmpeg: Denial of Service via crafted RTP/ASF stream CVE-2026-66036 — ffmpeg: FFmpeg: Arbitrary code execution via crafted video in vf_hqdn3d filter CVE-2026-66039 — ffmpeg: FFmpeg: Arbitrary code execution via crafted CAF file CVE-2026-69147 — vllm: vLLM: GPU memory exhaustion via PyNvVideoCodec decode bypass CVE-2026-71486 — vllm: vLLM: Denial of Service via unbounded token ID decoding CVE-2026-73066 — tesseract: Tesseract: Heap out-of-bounds write via crafted .traineddata CVE-2026-73072 — vim: Vim: Heap buffer overflow allows arbitrary code execution CVE-2026-73076 — vim: Vim: Arbitrary command execution via crafted vimball CVE-2026-73077 — vim: Vim: Arbitrary Code Execution via Insecure Shell Command Handling CVE-2026-73078 — vim: Vim: Arbitrary Code Execution via Crafted Netrw Menu Entries CVE-2026-73555 — vllm: vLLM: Information Disclosure via Validation Error Messages CVE-2026-73556 — vllm: vLLM: Denial of Service via Regular Expression processing CVE-2026-73557 — vllm: vLLM: Incomplete remediation allows bypass via concurrent prompt processing CVE-2026-73558 — vllm: vLLM: Cross-User Data Leakage via Integer Overflow CVE-2026-73559 — vllm: vLLM: Denial of Service via unbounded completion prompt lists CVE-2026-73560 — vllm: vLLM: Server-Side Request Forgery and arbitrary file read via improper media processing CVE-2026-84378 — httpx2: HTTPX2: Denial of Service via crafted Server-Sent Events stream CVE-2026-84379 — httpx2: HTTPX2: Multipart header injection via unvalidated input CVE-2026-84380 — httpx2: HTTPX2: Request Smuggling and Connection Desynchronization via Conflicting HTTP Headers CVE-2026-84381 — httpx2: httpcore2: HTTPX2: WebSocket traffic sent in plaintext via SOCKS5 proxy due to TLS failure CVE-2026-84382 — httpx2: HTTPX2: Denial of Service via streaming response decompression memory amplification CVE-2026-93592 — vllm: vLLM: Denial of Service via negative token ID input CVE-2026-97687 — urllib3: urllib3: Traffic interception via HTTPS proxy TLS configuration override CVE-2026-97688 — urllib3: urllib3: Denial of Service via infinite loop during chunked Deflate decoding CVE-2026-97689 — urllib3: urllib3: Denial of Service via unbounded memory allocation in chunk parser CVE-2026-101917 — pyjwt: PyJWT: Denial of Service via outbound request amplification on unknown key identifiers CVE-2026-101918 — pyjwt: PyJWT: Denial of Service via deeply nested JSON token payload CVE-2026-102265 — pyjwt: pyjwt: Denial of Service via deeply nested token headers CVE-2026-102266 — pyjwt: pyjwt: Authentication bypass via empty HMAC key acceptance CVE-2026-102267 — pyjwt: pyjwt: Verification key substitution via unvalidated JWKS redirects CVE-2026-102268 — pyjwt: PyJWT: Authentication bypass via improper public key validation CVE-2026-102269 — pyjwt: PyJWT: Token revocation bypass via non-canonical signature decoding CVE-2026-102270 — pyjwt: pyjwt: Denial of Service via regular expression backtracking in is_pem_format CVE-2026-102271 — pyjwt: PyJWT: Authentication bypass via acceptance of DER public keys as HMAC secrets CVE-2026-102272 — pyjwt: pyjwt: Token forgery via improper Unicode byte-order mark handling CVE-2026-102273 — pyjwt: pyjwt: Token forgery via acceptance of public JWK containers as HMAC secrets CVE-2026-102274 — pyjwt: pyjwt: Denial of Service via malformed RSA key in JWK set CVE-2026-102275 — pyjwt: PyJWT: Token verification bypass via mismatched OKP key components CVE-2026-102598 — werkzeug: werkzeug: Denial of Service via safe_join() Windows special device paths CVE-2026-103001 — pyjwt: pyjwt: Token validation bypass via options dictionary mutation CVE-2026-104874 — multidict: multidict: Denial of Service via memory leak in items-view operations

🎯 Affected products2

  • Red Hat AI Inference Server 3.5
  • registry.redhat.io/rhaii/vllm-gaudi-rhel9@sha256:ba46859d4cb085c111a27bc665c15e5f826320e86c03b940f7a28f40df79ad22_amd64 as a component of Red Hat AI Inference Server 3.5

✅ Remediation

For more information visit https://access.redhat.com/errata/RHSA-2026:82061 Workaround: Mitigation for this issue is either not available or the currently available options don't meet the Red Hat Product Security criteria comprising ease of use and deployment, applicability to a widespread installation base, or stability. It is strongly recommended to apply the upstream patch once available. Workaround: Mitigation for this issue is either not available or the currently available options do not meet the Red Hat Product Security criteria comprising ease of use and deployment, applicability to widespread installation base or stability. Workaround: Mitigation for this issue is either not available or the currently available options do not meet the Red Hat Product Security criteria comprising ease of use and deployment, applicability to widespread installation base, or stability. Workaround: No mitigation is currently available that meets Red Hat Product Security's standards for usability, deployment, applicability, or stability. Customers are advised to apply the appropriate security update once it becomes available. Workaround: Do not install a MIME magic file or content from untrusted sources. Workaround: Avoid running vLLM with python -O or PYTHONOPTIMIZE=1 until updated packages are available. Only load models from trusted sources. Restrict who can deploy or update models on inference endpoints. Apply network access controls and authentication in front of vLLM APIs. Workaround: To mitigate this issue, users should exercise caution when opening untrusted files or repositories with Vim, particularly those that might trigger the `cucumber` filetype plugin. Avoiding interaction with untrusted content can prevent the execution of malicious step-definition patterns. Workaround: Users can mitigate this vulnerability by disabling Python omni-completion in Vim if it is not essential for their workflow. This prevents the execution of untrusted Python code when opening hostile files. To disable this feature, ensure that the `omnifunc` option in your Vim configuration (e.g., `~/.vimrc`) is not set to `pythoncomplete#Complete` or `python3complete#Complete`. Alternatively, users should avoid invoking omni-completion (`Ctrl-X Ctrl-O`) on Python files from untrusted sources. Disabling Python omni-completion may affect Python development functionality within Vim. Workaround: Restrict unprivileged users from creating symlinks in directories that privileged processes operate on with ACL commands. Where possible, use the fs.protected_symlinks sysctl (enabled by default on RHEL 7+), which prevents symlink following in world-writable sticky directories unless the owner of the symlink matches the owner of the target file or directory. Workaround: To mitigate this issue, administrators should ensure that the `pam_userdb` module is not configured to store or compare credentials in plaintext. Verify that `pam_userdb` is either configured with a strong cryptographic hashing method or, if not required, is disabled. Avoid using `crypt=none` or omitting the `crypt=` argument when configuring `pam_userdb`. If changes are made to PAM configuration files, services relying on PAM may need to be restarted for the changes to take effect. Workaround: To mitigate this issue disable spell checking or avoid using SOFO-based spell files. This can be achieved globally by adding set nospell to your ~/.vimrc configuration file. Ensure your systems utilize standard UTF-8 encoding. This flaw is strictly confined to legacy 8-bit encodings and cannot be triggered under default Red Hat configurations. Workaround: To mitigate this vulnerability, users should avoid opening untrusted Python files or using Python omni-completion on such files. If Python omni-completion is not required, it can be disabled by adding `autocmd FileType python setlocal omnifunc=` to your `.vimrc` file. This will prevent the vulnerable code from being executed. Disabling Python omni-completion will remove the ability to use `Ctrl-X Ctrl-O` for Python code completion. A restart of Vim is required for the changes to take effect. Workaround: To mitigate this vulnerability, in applications processing user-supplied dates, implement input validation to ensure the supplied date is within the supported range before calling g_date_time_add_full() with untrusted data, specifically rejecting inputs that result in a negative or zero days field. Workaround: To mitigate this vulnerability, implement strict input validation to sanitize user-supplied replacement strings, specifically rejecting or escaping case-change modifiers (\u, \l, \U, \L) before calling g_regex_replace() or g_regex_replace_eval() when the G_REGEX_RAW compile flag is used. Removing the G_REGEX_RAW flag or hardcoding the replacement strings will completely neutralize this issue. Workaround: To mitigate this vulnerability, restrict any custom line terminator string passed to g_io_channel_set_line_term() to a maximum length of one byte before calling g_io_channel_read_line_backend(). Using the default line terminators will completely neutralize this issue. Workaround: To mitigate this vulnerability, implement input validation to sanitize untrusted key files (such as .desktop or .ini files), specifically rejecting or stripping empty values before calling g_key_file_get_locale_string_list(). Alternatively, restricting the application to only load key files from trusted sources will completely neutralize this issue. Workaround: To mitigate this vulnerability, ensure that applications only connect to trusted D-Bus servers and operate within secure, isolated networks to prevent man-in-the-middle (MitM) attacks. If feasible, configuring the D-Bus connection to strictly require the EXTERNAL authentication mechanism and disabling DBUS_COOKIE_SHA1 will completely neutralize this issue. Workaround: Users should exercise caution when opening untrusted PHP files and avoid invoking omni-completion on them. To prevent exploitation, the PHP omni-completion script can be disabled by moving or renaming `phpcomplete.vim`. For example, execute `mv /usr/share/vim/vim*/autoload/phpcomplete.vim /usr/share/vim/vim*/autoload/phpcomplete.vim.bak`. This action will disable PHP omni-completion functionality. A restart of Vim is necessary for this change to take effect. Workaround: Users are advised to avoid opening untrusted C source files or project tags files in Vim. Exercising caution and only processing trusted content prevents exploitation. Workaround: To mitigate this issue, OpenSSH clients should only connect to trusted SSH servers. Enforcing strict host key checking and carefully managing `known_hosts` files can help prevent connections to servers with unexpected or altered host keys, thereby reducing exposure to this client-side vulnerability. Workaround: To mitigate this issue, avoid processing VobSub subtitle files from untrusted or unknown sources. For applications that utilize FFmpeg's VobSub demuxer, consider running them within a sandboxed environment to restrict the potential impact of a successful exploit. Workaround: To mitigate this issue, restrict network access to applications utilizing FFmpeg for RTP/ASF stream demuxing, ensuring only trusted sources can provide such streams. Implement firewall rules to limit inbound connections to the affected services. This may impact legitimate functionality if trusted sources are inadvertently blocked. Workaround: To mitigate this issue, avoid processing `.traineddata` files from untrusted sources with Tesseract. Ensure that only `.traineddata` files from known, reputable origins are used for OCR recognition. Workaround: If spell checking is unused, disable it with set nospell in ~/.vimrc and do not set spelllang. If spell is required, load only trusted .spl files from Vim’s spell directories and do not place untrusted spell files on runtimepath. Additionally, consider disabling modelines (set nomodeline in ~/.vi…

🔗 References (75)