CVE-2026-49468: LiteLLM Authentication Bypass by Spoofing

CVE-2026-49468 is a critical authentication bypass by spoofing vulnerability in LiteLLM, published in the National Vulnerability Database (NVD) on June 22, 2026. It carries a CVSS v3.1 base score of 9.8 (Critical). According to its CVSS vector, it can be exploited over the network, needs no privileges and no user interaction.

What is CVE-2026-49468?

Host header flaw in LiteLLM proxy may allow unauthenticated access to management routes.

The NVD description reads:

LiteLLM is a proxy server (AI Gateway) to call LLM APIs in OpenAI (or native) format. Prior to 1.84.0, a Host-header parsing flaw in the LiteLLM proxy could, under specific conditions, allow unauthenticated access to protected management routes. The auth layer derived the effective route from request.url.path in litellm/proxy/auth/auth_utils.py::get_request_route(), which Starlette reconstructs from the Host header. A crafted Host could therefore make the auth gate evaluate a different route from the one FastAPI dispatched. This vulnerability is fixed in 1.84.0.

NVD maps this issue to CWE-290 (Authentication Bypass by Spoofing).

Affected versions and packages

NVD's CPE configuration lists these vulnerable version ranges:

• litellm before 1.84.0

Based on those ranges, the first versions that are no longer listed as vulnerable include: 1.84.0.

Echo's advisory data tracks 1 package for CVE-2026-49468: litellm.

Echo-built fixed versions recorded for this CVE:

• litellm: 1.82.0+echo.1

Other package versions listed in Echo's advisory data (compare them with the versions you run):

• litellm: 1.82.0, 1.82.6

Severity and CVSS score

A score of 9.8 puts CVE-2026-49468 in the Critical range (9.0 to 10.0), the highest CVSS band. The vector shows the attack path and the impact on confidentiality, integrity and availability. For help weighing exploit likelihood against raw severity, see our guide to managing Docker vulnerability risk.

Impact: Authentication Bypass by Spoofing

The CVSS vector rates the impact on confidentiality as high, integrity as high and availability as high. Exploitation requires no privileges and no user interaction, and the attacker reaches the target over the network.

What this means in practice depends on how LiteLLM is used in your environment: how widely it is deployed, whether the affected code path is reachable, and what the process can access. Treat the exposed systems that run a listed version as in scope until you have confirmed otherwise.

Is CVE-2026-49468 being exploited in the wild?

Known exploitation: CISA KEV

No. As of November 4, 2026, CVE-2026-49468 is not listed in CISA's Known Exploited Vulnerabilities (KEV) catalog (catalog version 2026.11.03). KEV only lists vulnerabilities with confirmed exploitation, so absence from the catalog is not proof that nobody is exploiting the flaw.

EPSS score and percentile

The Exploit Prediction Scoring System (EPSS) from FIRST estimates the probability that a CVE will see exploitation activity in the next 30 days. On November 3, 2026, CVE-2026-49468 had an EPSS score of 0.030, which corresponds to a probability of 3.0%. It sits at the 88th percentile, which means it is scored as more likely to be exploited than roughly 87% of scored CVEs, placing it above the median of scored CVEs. EPSS is recalculated daily and is a likelihood estimate, not a confirmation, so read it together with KEV status, exposure and CVSS severity.

Attack vector, complexity, privileges and user interaction

Based on the CVSS v3.1 vector CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H (source: NVD), this is how an attacker has to approach CVE-2026-49468:

• Attack vector: Network (AV:N): the flaw can be reached remotely over the network, so the attacker does not need to be on the same host or LAN.
• Attack complexity: Low (AC:L): no special conditions are needed, so the attack can be repeated reliably.
• Privileges required: None (PR:N): no account or credentials are needed.
• User interaction: None (UI:N): no action by a user is required.
• Scope: Unchanged (S:U): the impact stays within the security scope of the vulnerable component.

Is there a public proof of concept or exploit?

CISA's SSVC assessment recorded in NVD (dated June 23, 2026) lists the exploitation status as "none", which means there was no evidence of active exploitation and no public proof of concept at that time. None of the references in the NVD record is tagged as an exploit. This reflects NVD and CISA records as of November 4, 2026, and public exploit code can appear at any time.

Practical exposure conditions

• Only systems that run an affected version are exposed. Compare the LiteLLM versions in your hosts and container images against the vulnerable ranges listed above.
• The vector is network-based, so LiteLLM deployments that process input from untrusted networks carry the most risk, especially internet-facing ones. Internal-only deployments are still exposed to anyone who can reach them.
• No credentials are needed, so exposure does not depend on how well accounts are protected.
• No user has to click or open anything, so the flaw can be triggered without any user action.
• CISA's SSVC assessment marks the issue as automatable, which means exposed instances can be found and attacked at scale.

How to fix CVE-2026-49468

1. Inventory. Find every host, container image and base image that includes LiteLLM, including the packages listed above (litellm).
2. Upgrade. Upgrade to 1.84.0 or later where NVD lists a fixed release, or to the Echo-built fixed package listed above.
3. Rebuild and restart. Rebuild container images and restart the services that load LiteLLM so the patched code is actually used.
4. Verify. Re-scan your images and hosts to confirm that no package still matches a vulnerable version of CVE-2026-49468.

Interim mitigations if you cannot patch immediately

These are general hardening steps, not a vendor-specific workaround. Updating is the only complete fix.

• Restrict network access to services that use LiteLLM to trusted sources with firewall rules or allow-lists until you can patch.
• Monitor logs and runtime behavior on affected systems for unusual activity, and prioritize them in your next patch window.

How Echo helps with CVE-2026-49468

Echo is a registered CVE Numbering Authority (CNA) and ships patched, vetted open source packages and container images, so teams can remove vulnerable components instead of tracking an endless backlog. For this CVE, Echo's advisory data tracks 1 package (litellm) and includes Echo-built fixed versions. Read how automated patch SLAs reduce enterprise risk in how automated patch SLAs reduce enterprise risk and why how to deal with CVE backlogs matters. Related LiteLLM advisories: CVE-2026-37004.

Frequently asked questions about CVE-2026-49468

What is CVE-2026-49468?

CVE-2026-49468 is a critical authentication bypass by spoofing vulnerability in LiteLLM. Host header flaw in LiteLLM proxy may allow unauthenticated access to management routes. It has a CVSS base score of 9.8.

Which versions are affected by CVE-2026-49468?

NVD lists these vulnerable ranges: litellm before 1.84.0.

Is CVE-2026-49468 actively exploited?

It is not listed in CISA's KEV catalog as of November 4, 2026 (catalog version 2026.11.03), but that does not rule out exploitation. FIRST's EPSS model puts the probability of exploitation activity in the next 30 days at 3.0%, at the 88th percentile (November 3, 2026). On public exploit code: CISA's SSVC assessment in NVD (dated June 23, 2026) lists no public proof of concept.

How do I fix CVE-2026-49468?

Upgrade to 1.84.0 or later where NVD lists a fixed release, or to the Echo-built fixed package listed above. Then rebuild images, restart affected services and re-scan to confirm the fix.

How severe is CVE-2026-49468?

It is rated Critical with a CVSS base score of 9.8. The vector CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H means it can be exploited over the network, with no privileges and no user interaction.

Are there other LiteLLM vulnerabilities I should patch?

Yes. Echo's advisory data also covers CVE-2026-37004. Patch them together to reduce the number of upgrade cycles.