Vulnerability report

Exploitation Attempts Observed Confirmed Confidence Not in CISA KEV

CVE-2026-115886

SSRF on unauthenticated MLflow servers Server-Side Request Forgery

Mlflow · >= 3.12.0rc0, < 3.17.0

Watch CVE-2026-115886 Browse KEV feed
Severity
CVSS 6.5 · Medium
Confidence
Confirmed
Exploit status
Observed in sensors
EPSS
0.7%
First observed
28 Oct 2026
Last observed
29 Oct 2026

Decision summary

What security teams need to know first

Direct answers before the deeper technical record.

What it is

CVE-2026-115886 is an unauthenticated vulnerability affecting mlflow. MLflow is an open source AI engineering platform for agents, large language models, and machine learning models. From 3.12.0rc0 until 3.17.0, the...

Have exploitation attempts been observed?

Yes. Previdian sensors have observed 38 exploitation attempts, 7 unique attacker IPs, and 2 countries targeting CVE-2026-115886. First observed Oct 28, 2026; last observed Oct 29, 2026. Confidence is confirmed.

Who is affected?

mlflow >= 3.12.0rc0, < 3.17.0.

What should we do?

Patch immediately, validate internet-facing exposure, and monitor for matching requests.

Overview

MLflow is an open source AI engineering platform for agents, large language models, and machine learning models

MLflow is an open source AI engineering platform for agents, large language models, and machine learning models. From 3.12.0rc0 until 3.17.0, the scorer invoke, register, and GenAI evaluate endpoints, including unauthenticated POST /ajax-api/3.0/mlflow/scorer/invoke, accept serialized third-party scorer data that Scorer.model_validate_json resolves using a module-path allowlist without constraining the selected class or its constructor arguments. When optional trulens and trulens-providers-litellm packages are installed, an attacker can select mlflow.genai.scorers.trulens, choose a model URI outside MLflow's native provider registry so execution uses the LiteLLM fallback, and supply api_base and api_key values that cause the server to send an outbound request to an attacker-chosen or internal host with attacker-controlled headers and body.

The vulnerable deserialization occurs before a tracking-store lookup or job submission. This issue is fixed in version 3.17.0.

mlflow Affected
mlflow
>= 3.12.0rc0, < 3.17.0
Published
23 Oct 2026
Exploitation Attempts Reported
28 Oct 2026
Attack vector
Remote
Complexity
Low
Privileges
None
User interaction
None

Tags

nuclei_scanner

CVE References

Exploitation evidence

Evidence of exploitation attempts

Third-party attestation and first-party sensor observation are shown separately so teams can judge the evidence chain.

Exploitation Attempts Observed in the Wild

Previdian

Recorded 28 Oct 2026

Previdian sensors observed exploitation attempts targeting this vulnerability.

Exploitation Attempts Observed

Previdian sensor

First observed 28 Oct 2026

Previdian sensors observed exploitation attempts targeting this vulnerability.

Proof of concept available

GitHub Advisory Database

Recorded 23 Oct 2026

Public scanner or PoC coverage increases practical exploitability.

Known exploited vulnerability sources

Per-source evidence links for KEV attestations are available through the Previdian Pro API.

Learn about Pro API access
Source Added
Previdian Sensors First 2026-10-29 12:20 UTC

Operational indicators for this CVE are listed under Detection.

Sensor telemetry

Previdian Sensor Observations

Previdian sensor observations are exploitation attempts targeting this vulnerability. They do not show that the target was vulnerable or that the attempt succeeded. The cards and chart show volume, unique attackers, and daily activity.

38

Attempts observed

7

Unique attacker IPs

2

Attacker countries

IN · US

4

Sensors observed

CVE-2026-115886 exploitation attempts over the last 7 days

Daily events observed by Previdian sensors

Updated 30 Oct 2026

0
0
0
0
18
20
0
24 Oct 25 Oct 26 Oct 27 Oct 28 Oct 29 Oct 30 Oct

First observed 28 Oct 2026 · Last observed 29 Oct 2026

Pro adds sensor region and window summaries. Enterprise adds raw IPs, paths, User-Agents, and payloads.

Start 14-day Pro trial

Detection

Operational artifacts and observed signals

Make the evidence actionable in scanner, SOC, and edge-control workflows.

Observed signals

Request targets
User-Agents
Callback hosts
1
27
0

Request targets and User-Agents available in Pro. Callback host details available in Enterprise.

Virtual patch status

No Previdian virtual patch is currently available. Future rules ship for ModSecurity, Cloudflare, and AWS WAF.

Learn about virtual patches →

Attacker IP Indicators

Attacker IP indicators observed · available in Pro and Enterprise.

Sensor-derived attacker IP indicators are available to Pro and Enterprise accounts under Detection and through the Pro API.

Learn about Pro API access

Scanner Artifacts

Scanner and exploit-framework references linked to this CVE.

Scanner Reference Detected
Nuclei https://github.com/projectdiscovery/nuclei-templates/blob/main/http/cves/2026/CVE-2026-115886.yaml 29 Oct 2026

Risk and context

Severity, weaknesses, and research context

CVSS v3.1

6.5 Medium

Potential damage if exploited. Separate from whether attackers are using it.

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

EPSS

0.7%

Estimated chance of exploitation in the next 30 days. Previdian's warning comes from evidence, not this score.

Weaknesses

Use of Externally-Controlled Input to Select Classes or Code ('Unsafe Reflection') CWE-470 Server-Side Request Forgery (SSRF) CWE-918

Recent mention · Github Advisory Database (PIP)

[mlflow] MLflow: SSRF on unauthenticated MLflow servers via third-party scorer kwargs in the scorer invoke, register, and GenAI evaluate endpoints

Summary MLflow's tracking server exposes POST /ajax-api/3.0/mlflow/scorer/invoke with no authentication, and the handler reflectively instantiates an attacker-named class from attacker-supplied JSON with unconstrained constructor kwargs. With the optional trulens +...

Read full advisory

Potential Proof of Concepts

These PoCs are unverified and could contain malware. Use at your own risk.

MLflow <3.17.0 - Unauthenticated SSRF via Third-Party (TruLens) Scorer

nuclei · Created Unknown

https://github.com/mlflow/mlflow/security/advisories/GHSA-2wqp-q3c6-ccj5

ghsa · Created Unknown · AI assessment 99%

MLflow: SSRF on unauthenticated MLflow servers via third-party scorer kwargs in the scorer invoke, register, and GenAI evaluate endpoints

Timeline

From disclosure to observed exploitation attempts

  1. 04:31 UTC

    Nuclei template available

    Scanner coverage available

  2. 03:26 UTC

    Indicators of compromise added (7)

    Indicators of compromise recorded

  3. 03:26 UTC

    Observed by Previdian sensors

    Previdian sensors observed exploitation attempts targeting this vulnerability.

  4. 01:20 UTC

    Added to the Previdian watchlist

    Pro and Enterprise Watch users had 5 days of early warning before Previdian KEV.

  5. 22:10 UTC

    Public PoC available

    Public proof-of-concept code published

  6. 21:17 UTC

    CVE published

    Vulnerability disclosed publicly

  7. 19:37 UTC

    CVE ID reserved

    Identifier reserved by the CNA

Pro API

Automate this intelligence

Confidence, exploit status, sensor telemetry, PoCs, scanner integrations, mentions, and tags are available programmatically for VM, SOC, and CTI workflows.

Start 14-day Pro trial Read API docs
  • Evidence confidence and provenance
  • First-party sensor telemetry
  • PoC and scanner references
  • Affected versions and enrichment
  • Automation-ready JSON delivery

GET /api/v2/pro/kevs/CVE-2026-115886

Sample API response 
{
  "cve_id": "CVE-2026-115886",
  "confidence": "Confirmed",
  "cvss_score": 6.5,
  "cvss_estimated": false,
  "epss_score": 0.00676,
  "exploit_status": {
    "exploited_in_the_wild": true,
    "active_exploitation_observed": true
  },
  "sensor_telemetry": { "attempts": 38, "sensors": 4 }
}