CVE-2026-59822

BerriAI LiteLLM Improper Authentication Vulnerability

Actively exploited (CISA KEV) High 8.2 CVSS 3.1

Description

LiteLLM is a proxy server (AI Gateway) to call LLM APIs in OpenAI (or native) format. Prior to 1.84.0, LiteLLM's MCP Streamable HTTP endpoint allowed an unauthenticated attacker to use a fabricated Authorization header to trigger an OAuth2 passthrough fallback path that replaced failed LiteLLM key validation with an empty UserAPIKeyAuth() object, allowing requests to reach MCP tooling without a valid LiteLLM key. This issue is fixed in version 1.84.0.

Description as published by the source, in English.

Key facts

Published: 2026-07-08
Last modified: 2026-09-03
Assigned by: security-advisories@github.com
CVSS vector: CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:L/A:N
Scored by: nvd@nist.gov (NVD Primary)
Probability of exploitation: 0.2 % higher than 16 % of all known vulnerabilities
EPSS, next 30 days, model v2026.06.15
Fix available: Yes, see references

Actively exploited

CISA added this vulnerability to the Known Exploited Vulnerabilities catalogue on 2026-09-02, with a remediation deadline of 2026-09-16 for US federal agencies.

Required action: Apply mitigations in accordance with vendor instructions, ensuring compliance with CISA’s BOD 26-04 Prioritizing Security Updates Based on Risk (see URL in Notes) guidance and CISA’s “Forensics Triage Requirements” (see URL in Notes). Follow applicable BOD 26-04 guidance for cloud services or discontinue use of the product if mitigations are unavailable. Stakeholders are responsible for evaluating each asset's internet exposure and ensuring adherence to BOD 26-04 patching guidelines.

Affected products

Vendor	Product	Versions	Status
Litellm	Litellm	< 1.84.0	Affected

Weakness type

CWE-287: Improper Authentication
When an actor claims to have a given identity, the product does not prove or insufficiently proves that the claim is correct.
CWE-306: Missing Authentication for Critical Function
The product does not perform any authentication for functionality that requires a provable user identity or consumes a significant amount of resources.

Attack techniques

Derived through the weakness type: CVE to CWE to CAPEC to ATT&CK. This shows which techniques the weakness class is historically used for. It does not assert that this vulnerability has been exploited that way.

T1040: Network Sniffing credential-access, discovery
T1134: Access Token Manipulation stealth, privilege-escalation
T1185: Browser Session Hijacking collection
T1505.003: Web Shell persistence
T1548: Abuse Elevation Control Mechanism privilege-escalation
T1550.001: Application Access Token lateral-movement
T1557: Adversary-in-the-Middle credential-access, collection
T1563: Remote Service Session Hijacking lateral-movement

References

https://github.com/BerriAI/litellm/commit/73869f0faf7d11ee21adcb5f91b8c33a340b6c2c Patch
https://github.com/BerriAI/litellm/pull/26463 Issue Tracking, Patch
https://github.com/BerriAI/litellm/releases/tag/v1.84.0 Product, Release Notes
https://github.com/BerriAI/litellm/security/advisories/GHSA-7488-6r32-c95q Mitigation, Patch, Vendor Advisory
https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-59822 US Government Resource
https://www.wiz.io/blog/ai-infrastructure-honeypot Third Party Advisory

Related at Berigo

Articles

Seven new KEV entries, three of them in open developer components Mentions this CVE 2026-09-03
CISA added seven new vulnerabilities to its Known Exploited Vulnerabilities catalogue on 2 September 2026. Three of them sit in open components used …

Does this affect you?

A vulnerability matters only if you run the product, in an affected version, somewhere an attacker can reach.

Check your asset list Browse all vulnerabilities

Sources and freshness

CVE

What it is: A reference number naming one specific vulnerability, for example CVE-2024-9474.
What it tells you: Which vulnerability we are talking about, and in which product.
What it does NOT tell you: Nothing about how serious it is, whether anyone is exploiting it, or whether it concerns you.
Why it matters: Without a shared number, vendor, authority and tooling cannot discuss the same flaw.
Used with the others: The number is the key. NVD, KEV and EPSS all describe the same CVE.

NVD CWE CISA KEV EPSS

How to read this together

No single value should be used on its own. A high CVSS score does not mean a vulnerability is being exploited, and a low one does not mean it is harmless. The four sources answer four different questions.

CVE names the specific vulnerability.
NVD adds severity and technical detail about it.
CWE describes the type of mistake behind it.
CISA KEV shows that it is already being exploited.
EPSS estimates how likely exploitation is soon.

The order that decides what is urgent

Do you actually have the product?
Is the version you run affected?
Is it listed in CISA KEV?
Does it have a high EPSS score?
Is the system reachable from the internet or from others?
How important is the system to the business?
Are there compensating controls already in place?

The first two questions are not about the vulnerability at all. They are about you, and they remove most of the list before you start.

A short example

Fix this week

CVSS 6,5 (medium)

In CISA KEV: yes
EPSS: 94 per cent
A firewall facing the internet

Can be scheduled

CVSS 9,8 (critical)

In CISA KEV: no
EPSS: 0,04 per cent
A test server with no internet access

Sorted by CVSS alone, the one on the right comes first. It is the more serious flaw in theory. But the one on the left is being used against real organisations right now, and it sits where an attacker can reach it. That is the one to fix first.

The critical one still has to be fixed. It just does not have to be fixed today.

This product uses the NVD API but is not endorsed or certified by the NVD. CVE and the CVE logo are registered trademarks of The MITRE Corporation; CVE records are used under the CVE Terms of Use. CWE is a trademark of The MITRE Corporation, used per the CWE Terms of Use. KEV data from CISA (CC0). Exploitation probability from EPSS, a FIRST.org project. Open-source package data from OSV and the GitHub Advisory Database (CC-BY 4.0).

Scores and vectors are shown as published by the sources. The ordering of results is Berigo's own editorial assessment, described on the search page, and is not NVD data. An empty result never means a system is safe.