CASCACascade Atlas

Evidence dossier

CVE-2026-59822

LiteLLM: MCP Authentication Bypass via OAuth2 Passthrough Fallback

Exploited in the wild (CISA KEV since Sep 2, 2026). GitHub_M reports CVSS 4.0 8.8. EPSS estimates 0.8% exploit likelihood as of Oct 22, 2026.

63.8–68.0Priority evidence range
Evidence through Oct 22, 2026, 6:07 PM UTC

As of Oct 22, 2026

Normalized restatement

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.

State
PUBLISHED
Published
Jul 8, 2026
Updated
Sep 3, 2026
Evidence coverage
85%

Evidence chronology

What was known when

Download this view's receipt →
  1. Source dateFirst observed by CASCA
    CISA KEV

    Exploitation cataloged: BerriAI LiteLLM Improper Authentication Vulnerability

    Inspect raw assertion
    Field
    observed_exploitation
    Value
    BerriAI LiteLLM Improper Authentication Vulnerability
    Original evidence ↗
  2. Source dateSource date omittedFirst observed by CASCA
    CISA-ADP

    Record text: CISA ADP Vulnrichment

    Inspect raw assertion
    Field
    container
    Value
    CISA ADP Vulnrichment
    Original evidence ↗
  3. Source dateSource date omittedFirst observed by CASCA
    GitHub_M

    Record text: LiteLLM: MCP Authentication Bypass via OAuth2 Passthrough Fallback

    Inspect raw assertion
    Field
    container
    Value
    LiteLLM: MCP Authentication Bypass via OAuth2 Passthrough Fallback
    Original evidence ↗
  4. Source dateFirst observed by CASCA
    FIRST EPSS

    EPSS score: 0.84% probability · 57.29th percentile

    Inspect raw assertion
    Field
    model_probability
    Value
    Probability 0.008360000000; percentile 0.572920000000
    Original evidence ↗
ExploitationCatalog member

CISA KEV · catalog date Sep 2, 2026 · first observed Sep 2, 2026

Exploit likelihood0.84%

FIRST EPSS · score date Oct 22, 2026 · 57.3th percentile · first observed Oct 22, 2026

SeverityCVSS 8.8

GitHub_M · CVSS 4.0 · first observed Sep 3, 2026 · same-version scores align in this snapshot

Evidence detail

Source limits and decisions

Why each evidence state has this value

casca-unknown-reasons-v1
Exploitation statusLast verified source snapshot retained

The last verified source boundary remains in force for this claim.

Revision
evidence-policy-v1.1.0
Cutoff
Oct 22, 2026
Resolution
Review source contract
Exploit likelihoodEvidence supported

Eligible evidence is present for this bounded claim.

Revision
evidence-policy-v1.1.0
Cutoff
Oct 22, 2026
Resolution
None
Severity assessmentEvidence supported

Eligible evidence is present for this bounded claim.

Revision
casca-direct-cvss-eligibility-v1
Cutoff
Oct 22, 2026
Resolution
None
Affected productsSource-reported scope

The cited source assertion is retained while canonical product linkage remains open.

Revision
casca-factor-d-obligations-v1
Cutoff
Oct 22, 2026
Resolution
Resolve identity

Source comparison

Who said what

CISA KEVOriginal assertion
Exploitation cataloged

BerriAI LiteLLM Improper Authentication Vulnerability

Inspect raw assertion
Field
observed_exploitation
Value
BerriAI LiteLLM Improper Authentication Vulnerability
Source dateFirst observed by CASCAOriginal evidence ↗
CISA-ADPIndependent enrichment
Record text

CISA ADP Vulnrichment

Inspect raw assertion
Field
container
Value
CISA ADP Vulnrichment
Source dateSource date omittedFirst observed by CASCAOriginal evidence ↗
GitHub_MOriginal assertion
Record text

LiteLLM: MCP Authentication Bypass via OAuth2 Passthrough Fallback

Inspect raw assertion
Field
container
Value
LiteLLM: MCP Authentication Bypass via OAuth2 Passthrough Fallback
Source dateSource date omittedFirst observed by CASCAOriginal evidence ↗
FIRST EPSSOriginal assertion
EPSS score

0.84% probability · 57.29th percentile

Inspect raw assertion
Field
model_probability
Value
Probability 0.008360000000; percentile 0.572920000000
Source dateFirst observed by CASCAOriginal evidence ↗

Applicability

Cited product scope

Trace impact →
1Underlying assertions
1Canonical products
1Target assertions
0Constraint assertions

Grouped from 1 configuration nodes in this exact snapshot. Visual grouping is navigational; asset exposure and root cause require cited evidence.

Identity source boundaries
  • Cpe dictionary1,775,266 records · observed through 2026-07-21T06:45:29.809Z27d65b0f-b718-4b4f-bb79-c47c68d09dfa
  • Cpe match643,502 records · observed through 2026-07-21T08:13:17.697Z955dae73-7302-438b-aee1-058d7cc5d48e

2 scope groups

GitHub_M · source assertedBerriAIlitellmDirect source scope
Affected: < 1.84.0
Source-reported scope

Canonical linkage remains open; the cited source assertion is retained below.

Inspect raw assertion[{"status": "affected", "version": "< 1.84.0"}]
NVD CPE · APPLICATIONlitellmlitellmVulnerable target · 1 assertions
Any version (unconstrained) (< 1.84.0)Canonical identity product-5141e2b1adff7522ee8ab4f9b3544b79fcd27991c598d33af802ee49f3fcbba2Linked exact
Scope supported
Inspect raw assertion
  1. cpe:2.3:a:litellm:litellm:*:*:*:*:*:*:*:*
    Official link
    Linked exact
    Role
    Vulnerable target
    Configuration
    0 · node/0 · match 0
    Logic
    OR
    Version bounds
    through excluding 1.84.0
    Match ID
    0432af82-95fd-4358-9414-7b41889cefd2

Affected-product evidence

Accepted scope and product mapping

0 canonical links · 2 source-reported links

Source-reported scopeSource-reported scope

Canonical linkage remains open; the cited source assertion is retained below.

Vendor specified only by source · Product specified only by source

Source class
Direct cve affected
Assertions
1
Mapping revision
cpe23-exact-mapping-v1
Observed cutoff
Oct 22, 2026
Retained assertion IDs
0eb352f4-7f7c-434a-a023-4c51ee4d2b42
Source-reported scopeSource-reported scope

Canonical linkage remains open; the cited source assertion is retained below.

Vendor specified only by source · Product specified only by source

Source class
Nvd cpe vulnerable target
Assertions
1
Mapping revision
cpe23-exact-mapping-v1
Observed cutoff
Oct 22, 2026
Retained assertion IDs
c96b8d42-015b-4ccc-ad62-562fb9ed003a

Assessments

CVSS by origin

8.2
NVDCVSS 3.1 · role Primary · priority eligiblevalid_matchCVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:L/A:N
8.8
security-advisories@github.comCVSS 4.0 · role Secondary · priority eligiblevalid_matchCVSS:4.0/AV:N/AC:L/AT:N/PR:N/UI:N/VC:H/VI:L/VA:N/SC:N/SI:N/SA:N/E:X/CR:X/IR:X/AR:X/MAV:X/MAC:X/MAT:X/MPR:X/MUI:X/MVC:X/MVI:X/MVA:X/MSC:X/MSI:X/MSA:X/S:X/AU:X/R:X/V:X/RE:X/U:X
8.8
GitHub_MCVSS 4.0 · role unknown · display onlyDirect record assessment retained outside normalized eligibilityCVSS:4.0/AV:N/AC:L/AT:N/PR:N/UI:N/VC:H/VI:L/VA:N/SC:N/SI:N/SA:N

Direct CVE/CNA normalized decisions

8.8Priority eligible

GitHub_M

CVSS 4.0 · Primary · Original assertion · rank 1

CVSS:4.0/AV:N/AC:L/AT:N/PR:N/UI:N/VC:H/VI:L/VA:N/SC:N/SI:N/SA:N
Validation
Valid match
Recomputed
8.8
Decision reason
Evidence supported
Policy
casca-direct-cvss-eligibility-v1

Assessments are retained side by side under closed precedence. Cascade never averages CVSS.

Evidence boundaries

  • KEV membership is authoritative for the catalog, not proof of exposure in any environment.
  • EPSS is a dated model probability and not an individual-environment prediction.
  • Affected or fixed status applies only to the exact cited product and version scope; remaining scope stays source-scoped.
  • NVD-carried upstream facts remain derivative; independent corroboration requires a separately authored source.
  • Only NVD metrics validated under the generation-bound calculator are Public Priority eligible; direct CVE record metrics remain display-only.
  • NVD CVSS source eligibility is closed: NVD-authored, exact record-source, or registered same-CVE container origin; unmapped sources remain display-only.
  • Core replay supports the active and immediately prior generation; retrospective valid-at replay is deferred.
  • OSV aggregation and OSV-converted NVD material are not independent corroboration.
  • Red Hat facts are vendor assertions for the exact supplied products.
  • Nonmembership and not-yet-observed states are not proof of safety.