CVE-2026-59822
CVE-2026-59822 is a security flaw in BerriAI LiteLLM, a tool for building AI apps. Attackers can get into BerriAI LiteLLM without a valid password. The US Cybersecurity and Infrastructure Security Agency (CISA) added it to its catalog of flaws known to be used in attacks on 2 Sep 2026.[1]
BerriAI LiteLLM is business software or equipment. Most families won’t have it, but attackers break into it to get inside companies, and from there to what those companies hold about their customers, patients and staff. BerriAI has published a fix or advice.
Description
CISA calls it an improper authentication in BerriAI LiteLLM. In its words: “BerriAI LiteLLM contains an improper authentication vulnerability in the MCP Streamable HTTP endpoint that could allow an unauthenticated attacker to establish an authenticated MCP session using an arbitrary Bearer token.”[1]
The CVE record adds: “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.”[2]
The underlying weakness is improper authentication (CWE-287), missing authentication (CWE-306), in MITRE’s list of common software weaknesses.
It scores 8.2 out of 10 on the Common Vulnerability Scoring System, which rates it high, scored by NVD using CVSS 3.1.[2]
Affected versions
NVD’s record doesn’t list version ranges for it; BerriAI’s advisory has them[3]. CISA names the product as BerriAI LiteLLM.
Most families won’t have BerriAI LiteLLM; it’s for whoever looks after IT at work.
Exploitation
CISA added it to its catalog of known exploited vulnerabilities on 2 Sep 2026, 56 days after it was made public. CISA lists a flaw only when it has reliable evidence of it being exploited, and gave US federal agencies until 16 Sep 2026 to fix it.[1]
FIRST’s Exploit Prediction Scoring System, which estimates how likely a flaw is to be used in attacks in the next 30 days from what it sees across the internet, puts it at under 1%. That’s in the top 43% of the flaws it scores.[4]
CISA assesses it as automatable: attackers can run it against many targets at once.
CISA doesn’t say who was attacked, or how.
Fix
BerriAI has published a fix or advice.[3] CISA’s 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).”[1]
- Check whether you use it.BerriAI LiteLLM is business software or equipment, so most families won’t have it. If you run a business, or work from home on company equipment, ask whoever looks after your IT.
- Send this page to your IT person.Ask them to confirm it’s fixed, or not in use. CISA gave US government agencies until 16 Sep 2026 to fix it.
- If it can’t be fixed, take it offline.CISA’s advice when there’s no fix is to stop using the product, or take it off the internet, until there is one.
You can’t fix someone else’s server, but you can watch for what follows a break-in like this: a breach notice, a password reset you didn’t ask for, or a scam message that knows things it shouldn’t. HARD watches for those for you and your family.
Timeline
| Date | Event |
|---|---|
| The flaw is made public in the National Vulnerability Database, with a severity score of 8.2.[2] | |
| CISA adds it to its catalog of flaws known to be used in attacks.[1] | |
| CISA’s deadline for US government agencies to fix it.[1] | |
| FIRST puts the odds of it being used in attacks in the next 30 days at under 1%.[4] |
See also
Other BerriAI LiteLLM flaws
- CVE-2026-42271: command injection ·
- CVE-2026-42208: SQL injection ·
Other improper authentication flaws
- CVE-2026-105133, AhsayCBS: improper authentication ·
- CVE-2026-42018, JFrog Artifactory: improper authentication ·
- CVE-2026-82329, JFrog Artifactory: improper authentication ·
- CVE-2023-49105, ownCloud: improper authentication ·
- CVE-2026-65400, Apple macOS: improper authentication ·
Added the same day
- CVE-2026-9586, Sangoma Switchvox: SQL injection ·
- CVE-2026-83549, SonicWall SMA1000 Appliances: OS command injection ·
- CVE-2026-83548, SonicWall SMA1000 Appliances: server-side request forgery ·
- CVE-2026-82329, JFrog Artifactory: improper authentication ·
- CVE-2026-49869, Kestra OSS: OSS OS command injection ·
Every BerriAI flaw on CISA’s list · Everything added in September 2026
References
- ^ Cybersecurity and Infrastructure Security Agency. “BerriAI LiteLLM Improper Authentication Vulnerability”. Known Exploited Vulnerabilities Catalog, added 2 Sep 2026.
- ^ National Institute of Standards and Technology. “CVE-2026-59822 Detail”. National Vulnerability Database, published 8 Jul 2026.
- ^ BerriAI. Advisories and fixes: github.com.
- ^ FIRST. “Exploit Prediction Scoring System”. EPSS 0.008, percentile 57.7, 31 Oct 2026.
- CVE Program. “CVE-2026-59822”. The official CVE record.
This article is built from CISA’s catalog (version 2026.10.30), NVD’s record and FIRST’s scores, kept in HARD’s database and checked every morning. The plain English is HARD’s.