CVE WatchOct 22, 2026

CVE-2026-114612

LiteLLM is a proxy server (AI Gateway) to call LLM APIs in OpenAI (or native) format. From 1.87.0 until 1.100.4, 1.101.3, 1.102.2, 1.103.1,

CVSS

9.9

Critical

EPSS

0.2%

p14

KEV

—

Exploit Today

0-100

Published: Oct 22, 2026 · Last modified: Oct 22, 2026 · CWE-269 · CWE-345 · CWE-441

EPSS · 30d

Not enough EPSS history yet.

Technical description

LiteLLM is a proxy server (AI Gateway) to call LLM APIs in OpenAI (or native) format. From 1.87.0 until 1.100.4, 1.101.3, 1.102.2, 1.103.1, and 1.104.0-rc.2, the proxy reuses one cryptographic context for at-rest secrets and session tokens, allowing an authenticated internal user to have crafted metadata containing a forged administrator credential encrypted and returned by the API-key flow. Replaying that ciphertext as a bearer token causes the proxy to trust the forged proxy administrator identity, permitting administrative access and arbitrary commands through the MCP stdio endpoint. Versions 1.87.0 through 1.90.x require EXPERIMENTAL_UI_LOGIN to be enabled, while versions after 1.91.0 are exploitable in the default configuration. This issue was fixed in LiteLLM 1.100.4, 1.101.3, 1.102.2, 1.103.1, and 1.104.0-rc.2.

Official references

Related CVEs

CVECVSSEPSSKEVRExploitTitleMod.

CVE-2026-554597.4 HIG

—

———The Goobi viewer is a web application that allows digitised material to be displayed in a web browser. From version 25.01 to before version 26.05, AuthenticationEndpoint.verifyOpenIdToken builds its verifier as JWT.require(algorithm).withIssuer(issuer).build() and does not enforce the aud claim during cryptographic validation. The aud check is performed manually afterwards via a single string-equality call, which only handles single-string aud values. In a multi-RP IdP setup, a token whose single-string aud matches the viewer's client ID but was issued for a different relying party passes the check and authenticates the attacker as the token's sub. This issue has been patched in version 26.05.4h CVE-2026-568328.8 HIG

—

———PraisonAI is a multi-agent teams system. From version 4.5.2 to before version 4.6.59, praisonai.bots.DiscordApproval approves a pending dangerous tool call when it sees any later non-bot message in the configured Discord channel whose text is classified as approval, such as yes. As a result, any user who can post in the configured approval channel can approve a pending high-risk tool call by sending yes after the approval message appears. the same PoV also shows that the Slack and Telegram messaging approval backends have no explicit approver allowlist parameter, but the primary report-grade issue is the Discord backend's unthreaded channel cross-talk: the approving message does not need to be a reply or otherwise request-bound. This issue has been patched in version 4.6.59.8h CVE-2026-257849.8 CRI

26.9%

——8Initialization of a Resource with an Insecure Default vulnerability in Hallo Welt! GmbH BlueSpice (container image bluespice/wiki) allows privilege escalation. The default value of $mwsgTokenAuthenticatorServiceCIDR in the container's LocalSettings.php was intended to restrict internal service authentication to the container network, but evaluated to an empty string when the environment variable WIKI_SERVICE_TOKEN_AUTH_ALLOWED was not set. An empty value disables the IP restriction in the token authenticator component (mwstake/mediawiki-component-token-authenticator). As a result, any user with read access to the wiki (anonymous users on publicly readable wikis) could obtain an application token from the REST API and use it to get a session with sysop privileges as the internal service user. This issue affects BlueSpice: from 5.2.3 before 5.2.7, and the 5.2.7 container images published before 2026-10-19 15:37 UTC.11h CVE-2026-942688.1 HIG

7.0%

——2The Advanced Custom Fields: Extended WordPress plugin before 0.9.2.8, Advanced Custom Fields: Extended PRO WordPress plugin before 0.9.2.8 does not resolve the target of the update-user action in its front-end Forms module consistently between the stage that authorises the request and the stage that carries it out, which allows unauthenticated visitors to bypass the guard protecting administrator accounts, take over the site administrator's account and be issued a valid session for it. Exploitation requires the site to publish two separate front-end form configurations whose targets are mapped to visitor-submitted values.6h CVE-2026-1119827.2 HIG

2.5%

——1The User Profile Builder WordPress plugin before 4.0.5 does not verify that a user is authorized to edit the specific account selected on its front-end Edit Profile form, allowing an Administrator of a Multisite subsite to change any other user's account email, including a Network Super Admin's, and take over that account through the password-reset flow.6h CVE-2026-1105103.5 LOW

0.7%

——0The Essential Addons for Elementor WordPress plugin before 6.8.7 does not validate the HTML tag setting of some of its widgets before using it to build markup, which could allow users with the author role and above to inject iframe and form elements into published pages.6h