CVE-2026-93355: LiteLLM: JWT email fallback enables admin account takeover

HIGHCISA: ATTEND
Published September 28, 2026
CISO Take

LiteLLM's JWT authentication flow falls back to looking users up by email without checking the email_verified claim, so any attacker with a valid token from the configured identity provider can claim a victim's account, including proxy_admin, and overwrite the victim's stored identity binding to keep that access. The CVSS is 8.1 (network, low complexity, low privileges, no user interaction). The EPSS is low (0.00273, though the 82nd percentile), it is not in CISA KEV, no public exploit or Nuclei template exists, and CISA SSVC says ATTEND. The impact is high because administrative endpoints expose API keys for upstream LLM providers and user management, which makes this a credential and spend-control problem for any gateway that centralizes model access. Prioritize if you run LiteLLM Proxy with JWT/SSO auth and an IdP that allows self-registration or unverified emails: apply the fixed release from the Ox Security and VulnCheck advisories, enforce email verification at the IdP, and rotate the provider keys managed by the proxy if you suspect abuse.

Sources:NVDEPSSATLASox.securityvulncheck.com

What is the risk?

Risk is MEDIUM-HIGH, not critical. Exploitation is easy once the precondition is met (a valid IdP token carrying an unverified email that matches a victim). That precondition limits exposure to deployments with JWT auth enabled and an IdP that issues tokens for unverified emails, such as one with open self-registration. Impact is high for confidentiality and integrity (admin takeover, API key exposure, persistent identity rebinding). There is no availability impact. Exploitation signals are low: no KEV listing, no public exploit, EPSS 0.27%. Even so, the 82nd percentile and the persistence mechanism justify prompt patching. Affected versions and the fixed release were not provided in the data, so confirm them in the vendor advisory.

How does the attack unfold?

Initial Access
Attacker obtains a valid IdP JWT carrying an unverified email that matches a victim admin's address.
AML.T0012
Exploitation
LiteLLM's email-based fallback lookup matches the victim account without checking email_verified and grants its role.
AML.T0091.000
Persistence
The victim's stored identity binding is overwritten with the attacker's identity, so access survives later logins.
Impact
Attacker uses proxy_admin endpoints to read API keys and manage users, exposing provider credentials.
AML.T0055

What systems are affected?

PackageEcosystemVulnerable RangePatched
LiteLLMpip—No patch
63.3KOpenSSF 5.86 dependentsPushed 3d ago51% patched~49d to patchFull package profile →

Do you use LiteLLM? You're affected.

How severe is it?

CVSS 3.1
8.1 / 10
EPSS
0.3%
chance of exploitation in 30 days
Higher than 20% of all CVEs
Exploitation Status
Exploit Available
Exploitation: MEDIUM
Sophistication
Trivial
Exploitation Confidence
medium
○CISA SSVC: Public PoC
Composite signal derived from CISA KEV, VulnCheck KEV, CISA SSVC, EPSS, Metasploit, Exploit-DB, trickest/cve, Nuclei templates, and inthewild.io exploitation reports.

What is the attack surface?

AVACPRUISCIA
AVNetwork
ACLow
PRLow
UINone
SUnchanged
CHigh
IHigh
ANone

What should I do?

1 step
  1. 1) Upgrade LiteLLM to the fixed release listed in the Ox Security and VulnCheck advisories (check the vendor changelog; the exact version was not given in the available data). 2) Until patched, disable JWT authentication on the proxy or the email-based user lookup if the configuration allows it. 3) At the IdP, require verified emails and disable open self-registration or untrusted federated providers. 4) Review the stored identity bindings (user IDs or subjects) of admin and privileged accounts for unexpected changes. 5) Audit proxy logs for JWT logins where the token's subject differs from the stored binding, or where email_verified is false or absent. 6) Rotate the upstream provider keys and virtual keys managed by the proxy if you find signs of abuse. 7) Restrict access to the admin endpoints by network (VPN or allowlist).

What does CISA's SSVC say?

DecisionAttend
Exploitationpoc
AutomatableNo
Technical Impacttotal

Source: CISA Vulnrichment (SSVC v2.0). Decision based on the CISA Coordinator decision tree.

How is it classified?

Which compliance frameworks are affected?

This CVE is relevant to:

EU AI Act
Art. 15-Accuracy, robustness and cybersecurity
ISO 42001
A.6.2.6-AI system operation and monitoring
NIST AI RMF
MANAGE 2.3-Response to unknown or newly identified risks
OWASP LLM Top 10
LLM06-Excessive Agency

Frequently Asked Questions

What is CVE-2026-93355?

LiteLLM's JWT authentication flow falls back to looking users up by email without checking the email_verified claim, so any attacker with a valid token from the configured identity provider can claim a victim's account, including proxy_admin, and overwrite the victim's stored identity binding to keep that access. The CVSS is 8.1 (network, low complexity, low privileges, no user interaction). The EPSS is low (0.00273, though the 82nd percentile), it is not in CISA KEV, no public exploit or Nuclei template exists, and CISA SSVC says ATTEND. The impact is high because administrative endpoints expose API keys for upstream LLM providers and user management, which makes this a credential and spend-control problem for any gateway that centralizes model access. Prioritize if you run LiteLLM Proxy with JWT/SSO auth and an IdP that allows self-registration or unverified emails: apply the fixed release from the Ox Security and VulnCheck advisories, enforce email verification at the IdP, and rotate the provider keys managed by the proxy if you suspect abuse.

Is CVE-2026-93355 actively exploited?

No confirmed active exploitation of CVE-2026-93355 has been reported, but organizations should still patch proactively.

How to fix CVE-2026-93355?

1) Upgrade LiteLLM to the fixed release listed in the Ox Security and VulnCheck advisories (check the vendor changelog; the exact version was not given in the available data). 2) Until patched, disable JWT authentication on the proxy or the email-based user lookup if the configuration allows it. 3) At the IdP, require verified emails and disable open self-registration or untrusted federated providers. 4) Review the stored identity bindings (user IDs or subjects) of admin and privileged accounts for unexpected changes. 5) Audit proxy logs for JWT logins where the token's subject differs from the stored binding, or where email_verified is false or absent. 6) Rotate the upstream provider keys and virtual keys managed by the proxy if you find signs of abuse. 7) Restrict access to the admin endpoints by network (VPN or allowlist).

What systems are affected by CVE-2026-93355?

This vulnerability affects the following AI/ML architecture patterns: LLM gateway/proxy, model serving, multi-provider LLM routing, agent frameworks, enterprise SSO-integrated AI platforms.

What is the CVSS score for CVE-2026-93355?

CVE-2026-93355 has a CVSS v3.1 base score of 8.1 (HIGH). The EPSS exploitation probability is 0.27%.

What is the AI security impact?

Affected AI Architectures

LLM gateway/proxymodel servingmulti-provider LLM routingagent frameworksenterprise SSO-integrated AI platforms

MITRE ATLAS Techniques

AML.T0012Valid Accounts
AML.T0049Exploit Public-Facing Application
AML.T0055Unsecured Credentials
AML.T0091.000Application Access Token

Compliance Controls Affected

EU AI Act: Art. 15
ISO 42001: A.6.2.6
NIST AI RMF: MANAGE 2.3
OWASP LLM Top 10: LLM06

What are the technical details?

Original Advisory

LiteLLM contains a weak authentication vulnerability that allows an attacker holding a valid JWT from the configured identity provider to authenticate as any existing user by exploiting an email-based fallback lookup in the JWT authentication flow without verifying the email_verified claim. Attackers can present a token with an unverified email address matching a victim's account to inherit the victim's role, including proxy_admin privileges, and permanently overwrite the victim's stored identity binding to retain persistent unauthorized access to administrative endpoints exposing API keys and user management.

Exploitation Scenario

An attacker registers an account at the organization's IdP, or at a permissive federated provider it trusts, using the email address of the LiteLLM proxy_admin without completing verification. They authenticate to get a valid JWT with that unverified email and present it to the LiteLLM proxy. The email-based fallback matches the admin's account, assigns its proxy_admin role and overwrites the stored identity binding with the attacker's. The attacker then calls the administrative endpoints to list or create API keys, exfiltrate the upstream LLM provider credentials, modify routing and budgets, and keep persistent access while the legitimate admin is locked out.

Weaknesses (CWE)

CWE-1390 — Weak Authentication: The product uses an authentication mechanism to restrict access to specific users or identities, but the mechanism does not sufficiently prove that the claimed identity is correct.

Source: MITRE CWE corpus.

CVSS Vector

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

Timeline

Published
September 28, 2026
Last Modified
September 30, 2026
First Seen
September 28, 2026

Related Vulnerabilities