8 vulnerabilities across 6 products scored HIGH or above on September 21, 2026.

  • 🟠 HIGH: 8

Exploit Status Upgrades

The following CVEs from previous bulletins have been upgraded based on new exploit intelligence:

  • [UPGRADED] CVE-2026-93603 (patriksimek/vm2) — F1: exploitable → functional, AAS: 12.7 → 14.7 (CRITICAL → CRITICAL). Originally in 2026-09-18 bulletin.
  • [UPGRADED] CVE-2026-93605 (patriksimek/vm2) — F1: exploitable → functional, AAS: 12.7 → 14.7 (CRITICAL → CRITICAL). Originally in 2026-09-18 bulletin.
  • [UPGRADED] CVE-2026-84383 (strukturag/libheif) — F1: exploitable → functional, AAS: 10.9 → 12.9 (HIGH → CRITICAL). Originally in 2026-09-18 bulletin.
  • [UPGRADED] CVE-2026-12793 (jetmonsters/jetformbuilder_—_dynamic_blocks_form_builder) — F1: exploitable → functional, AAS: 10.1 → 12.1 (HIGH → CRITICAL). Originally in 2026-09-16 bulletin.

🟠 [HIGH] 1panel-dev/maxkb

2 CVEs | CVSS 3.1: 10.0 | AAS 11.2

  • cpe:2.3:a:1panel-dev:maxkb:*:*:*:*:*:*:*:* (< 2.10.5-lts)

1panel-dev MaxKB, an open-source AI assistant platform for enterprises, is affected by 2 vulnerabilities including at least one rated CVSS 10.0. The most severe issue allows unauthenticated command execution through the SandboxShellBackend, where assistants with tools, MCP tools, skills, or sub-applications can be manipulated via untrusted chat or ingested content to run arbitrary shell commands without requiring human approval. Source deployments with sandboxing disabled are especially at risk, as commands execute directly as the application user.

Organizations running MaxKB should upgrade to version 2.10.5-lts or later immediately. Teams using MaxKB in any deployment configuration, whether containerized or source-installed, should treat this as a critical priority given the trivial exploitability and full system compromise potential. Review the vendor commit at the linked advisory for patch details.

Vendor Advisory

🟠 [HIGH] ajenti/ajenti

1 CVE | CVSS 3.1: 9.9 | AAS 11.1

  • cpe:2.3:a:ajenti:ajenti:*:*:*:*:*:*:*:*

Ajenti, a modular server administration panel for Linux and BSD systems, is affected by 1 vulnerability rated CVSS 9.9. Any authenticated user, regardless of privilege level, can enqueue plugin management tasks without proper authorization, and because user-supplied package names and versions are passed to pip unsanitized while running as root, this enables full system compromise through arbitrary code execution.

Administrators running Ajenti should upgrade to version 2.2.16 or later immediately. Given that any authenticated account can exploit this flaw to gain root access, organizations should also audit recent plugin installation activity and restrict panel access until the patch is applied.

Vendor Advisory

🟠 [HIGH] apache_software_foundation/apache_mina

1 CVE | CVSS 3.1: 9.8 | AAS 9.9

  • cpe:2.3:a:apache:apache_mina:*:*:*:*:*:*:*:*

Apache MINA, a widely used network application framework, is affected by 1 vulnerability rated CVSS 9.8. A previous deserialization filter bypass fix was only applied to the 2.2.x branch, leaving the 2.0.x and 2.1.x maintenance branches still vulnerable despite being listed as patched, meaning organizations running MINA 2.0.29 through 2.0.30 or 2.1.13 through 2.1.14 remain exposed to remote code execution via proxy class deserialization.

Teams using Apache MINA on the 2.0.x or 2.1.x branches should migrate to the 2.2.x line or apply the fix once available for their branch, as the previously announced patches for those versions were incomplete. Review the vendor advisory on the Apache mailing list for updated guidance and verify which branch your deployments are actually running.

Vendor Advisory

🟠 [HIGH] feast-dev/feast

1 CVE | CVSS 4.0: 8.9 | AAS 9.2

  • cpe:2.3:a:feast-dev:feast:*:*:*:*:*:*:*:*

Feast, an open-source feature store for AI and machine learning, is affected by 1 vulnerability rated CVSS 8.9. A flaw in the GitHub Actions CI workflow allows fork contributors to execute arbitrary code with access to privileged cloud credentials for GCP, AWS, and Snowflake by obtaining approval on a benign commit and then pushing malicious changes that retain the approval labels.

Organizations using or contributing to Feast should upgrade to version 0.65.0 or later. Teams that maintain forks or accept external pull requests should also audit CI workflow logs for any unauthorized credential access and rotate any exposed cloud service credentials as a precaution.

Vendor Advisory

🟠 [HIGH] vllm-project/vllm

2 CVEs | CVSS 4.0: 8.7 | AAS 9.1

  • cpe:2.3:a:vllm:vllm:*:*:*:*:*:*:*:* (< 0.29.1)
  • cpe:2.3:a:vllm-project:vllm:*:*:*:*:*:*:*:*

vLLM, a widely adopted open-source LLM inference engine, is affected by 2 vulnerabilities including at least one rated CVSS 8.7. The most severe issue allows remote attackers to supply unvalidated parameters through OpenAI-compatible API endpoints to trigger unbounded memory allocation, causing kernel OOM-kill of worker processes and denial of service in disaggregated prefill/decode deployments.

Organizations running vLLM through version 0.29.0, particularly those exposing completion endpoints to untrusted users, should monitor the vendor repository for a patched release and apply it as soon as available. In the interim, teams should consider restricting API access to trusted clients and implementing input validation or resource limits at the network layer to mitigate exploitation.

Vendor Advisory

🟠 [HIGH] mailu/mailu

1 CVE | CVSS 3.1: 9.8 | AAS 9.1

  • cpe:2.3:a:mailu:mailu:*:*:*:*:*:*:*:*

Mailu, a Docker-based mail server suite, is affected by 1 vulnerability rated CVSS 9.8. Deployments from version 2.0 through 2024.06.55 that have PROXY_AUTH_WHITELIST configured without setting REAL_IP_HEADER trust a client-controlled X-Forwarded-By header for proxy authentication, allowing unauthenticated attackers to bypass authentication entirely by injecting the header in requests.

Organizations running affected Mailu versions should upgrade to 2024.06.55 or later, and Helm chart users should update to version 2.7.3 or later. Teams should also verify their configuration explicitly sets REAL_IP_HEADER and audit access logs for any signs of unauthorized authentication bypass via forged headers.

Vendor Advisory