CVE-2026-77521

CRITICAL10.0

Published 2026-09-21 · Updated 2026-09-21 · Source security-advisories@github.com

Description

MaxKB is an open-source AI assistant for enterprise. Prior to version 2.10.5-lts, assistants with a tool, MCP tool, skill, or sub-application use SandboxShellBackend, which exposes an execute shell tool without excluding it and omits execute from interrupt_on, so human approval is not required. Untrusted chat or ingested content can therefore cause command execution; source deployments with MAXKB_SANDBOX disabled run commands directly as the application user, while the official root container's string-based gosu wrapper allowed shell metacharacters to execute outside the intended sandbox. This issue is fixed in version 2.10.5-lts.

CVSS vector: CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:C/C:H/I:H/A:H

CWE-78CWE-250CWE-749

References

View on NVD

Related CVEs

Same severity, most recent

CVE-2026-79916
MaxKB is an open-source AI assistant for enterprise. Prior to 2.10.5-lts, authenticated workspace members can inject control characters into AWS Bedrock access_key_id and secret_access_key fields that _update_aws_credentials writes to /root/.aws/credentials without safe parsing. An attacker can append a new AWS profile containing credential_process, then select that profile during a later model-validation request so botocore executes an attacker-controlled command as root. This vulnerability is fixed in 2.10.5-lts.
2026-09-21 · score 9.1
CVE-2026-58491
Warpgate is an open source SSH, HTTPS and MySQL bastion host for Linux. Prior to 0.25.5, the /@warpgate/api/sso/providers/:name/start endpoint stores an attacker-controlled next parameter that the POST /@warpgate/api/sso/return handler inserts without HTML escaping into the response generated by warpgate-protocol-http/src/api/sso_provider_list.rs. A victim who follows a crafted link and completes SSO can cause markup and JavaScript to execute in the authenticated Warpgate origin, allowing access to session data and actions through user APIs, and through administrator APIs only when the victim is an administrator. The GET /@warpgate/api/sso/return path also uses the same unvalidated value as a redirect destination, enabling an open redirect. This issue is fixed in version 0.25.5.
2026-09-21 · score 9.3
CVE-2026-79920
Ajenti is a Linux & BSD modular server admin panel. Prior to version 2.2.16, any authenticated user can call /api/core/tasks/start to enqueue InstallPlugin, UnInstallPlugin, or UpgradeAll from plugins/plugins/tasks.py without plugin-management authorization. InstallPlugin and UnInstallPlugin construct a pip package specification from unvalidated name and version fields, and the task worker invokes pip while running as root. A low-privileged user can therefore select or manipulate a package installed with root privileges and can install, remove, or upgrade plugins without administrative permission, resulting in root code execution and full host compromise. This issue is fixed in version 2.2.16.
2026-09-21 · score 9.9
CVE-2026-85751
Mailu is a mail server distributed as a set of Docker images. From Mailu 2.0 until 2024.06.55 and prior to Mailu helm-charts 2.7.3, deployments with PROXY_AUTH_WHITELIST configured but REAL_IP_HEADER unset trusted a client-controlled X-Forwarded-By header for header-based proxy authentication. The proxy_hide_header directive in the nginx template at core/nginx/conf/proxy.conf hid the header from upstream responses but did not overwrite the incoming request value in this configuration. An unauthenticated remote attacker could therefore spoof the trusted proxy identity and bypass authentication. This issue is fixed in Mailu 2024.06.55 and Mailu helm-charts 2.7.3.
2026-09-21 · score 9.8
CVE-2026-94301
The fix for CVE-2026-47065/ZDRES-232 ("resolveProxyClass Not Overridden - acceptMatchers Filter Bypass via java.lang.reflect.Proxy"), released on 2026-06-02 and announced as "Fully addressed" in MINA 2.2.8, 2.1.13 and 2.0.29, was committed to the 2.2.X branch only. The 2.0.X and 2.1.X maintenance branches never received the resolveProxyClass() override, so the 2.0.29 and 2.1.13 artifacts listed as fixed -- and every later release on those lines, up to and including the current 2.0.30 and 2.1.14 -- remain vulnerable to the exact allow-list bypass that CVE-2026-47065 was meant to close.
2026-09-21 · score 9.8