VulnerabilityPublished 27 Oct 2026

CVE-2026-116691: MLflow CrewAI autologging records LLM provider API keys in logged trace spans

Affects mlflowSource GCVE Database

Authoritative record:CVE-2026-116691· Medium 6.5

Not known to be exploited · CVSS 6.5 Medium : upgrade to 3.17.0 when convenient.

Data as of 27 Oct 2026, 22:27 UTC

CVSS 3.1 base score

6.5Medium

How severe the flaw is on the 0–10 CVSS scale.

Patch: Available – fixed in 3.17.0

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

Attack Vector: Network
Attack Complexity: Low
Privileges Required: Low
User Interaction: None
Scope: Unchanged
Confidentiality: High
Integrity: None
Availability: None

Bold values are the worst case: easiest to exploit or largest impact.

Attack Vector: Network: The vulnerable component is bound to the network stack and the set of possible attackers extends beyond the other options listed, up to and including the entire Internet. Such a vulnerability is often termed 'remotely exploitable' and can be thought of as an attack being exploitable at the protocol level one or more network hops away (e.g., across one or more routers).
Attack Complexity: Low: Specialized access conditions or extenuating circumstances do not exist. An attacker can expect repeatable success against the vulnerable component.
Privileges Required: Low: The attacker is authorized with (i.e., requires) privileges that provide basic user capabilities that could normally affect only settings and files owned by a user. Alternatively, an attacker with Low privileges may have the ability to cause an impact only to non-sensitive resources.
User Interaction: None: The vulnerable system can be exploited without any interaction from any user.
Scope: Unchanged: An exploited vulnerability can only affect resources managed by the same security authority. In this case, the vulnerable component and the impacted component are either the same, or both are managed by the same security authority.
Confidentiality: High: There is total loss of confidentiality, resulting in all resources within the impacted component being divulged to the attacker. Alternatively, access to only some restricted information is obtained, but the disclosed information presents a direct, serious impact.
Integrity: None: There is no loss of integrity within the impacted component.
Availability: None: There is no impact to availability within the impacted component.

Definitions: FIRST CVSS v3.1 specification.

Exploitation evidence

Disclosed: evidence found.
CVE published 27 Oct 2026
Detectable: no evidence on record.
No public nuclei template indexed
Weaponised: no evidence on record.
No public exploit code indexed in Metasploit or Exploit-DB
Exploited in the wildExploited: no evidence on record.
Not in CISA KEV and no source reports exploitation
Observed by OffSeq MirageSeen by Mirage: no evidence on record.
No attack traffic recorded by OffSeq Mirage sensors

A missing step is not evidence of safety. Sources: cve.org (opens in a new tab), the Metasploit, Exploit-DB and nuclei-templates index as of 26 Oct 2026.

Disclosed: The CVE record is public on cve.org. Its publication is t = 0.
Detectable: A public nuclei template can find exposed instances, for defenders and attackers alike.
Weaponised: Public exploit code exists in Metasploit or Exploit-DB, or CISA SSVC or the CVSS exploit maturity reports a public proof of concept.
Exploited in the wild: CISA KEV lists it, CISA SSVC rates exploitation active, or a source reports attacks.
Observed by OffSeq Mirage: OffSeq Mirage honeypot sensors recorded attack traffic for this CVE, or found exposed hosts without attack traffic.

Moments are spaced evenly, not to scale; a dashed line marks a gap of six months or more. Deltas count from t = 0. Sources that give only a date (CISA KEV, Exploit-DB, or a time of exactly 00:00Z) read “same day” rather than hours. Metasploit modules carry the flaw’s disclosure date, not their release date, and nuclei templates carry no date, so both appear above but not on the timeline.

What to do

Upgrade mlflow to 3.17.0.

Affected software

Product	Affected	Fixed in
PyPImlflow	Affected >=2.19.0 <3.17.0	Fixed in 3.17.0

Check your own hosts for these packages with threat-finder (opens in a new tab), our free open-source scanner.

AI-suggested

A fix is available in MLflow version 3.17.0. Users should upgrade to MLflow 3.17.0 or later to prevent API key leakage. Until upgrading, avoid using CrewAI autologging with MLflow versions >=2.19.0 and <3.17.0 or disable telemetry export to untrusted endpoints. The vendor advisory indicates the issue is fixed; no additional mitigation is required beyond applying the official patch.

Description

MLflow's CrewAI autologging feature in versions >=2.19.0 and <3.17.0 leaks CrewAI LLM provider API keys in trace telemetry. The vulnerability arises because MLflow serializes the CrewAI Agent object's __dict__ including the LLM object, whose string representation contains the API key in plaintext. This sensitive information is unintentionally sent to MLflow's trace backend or any configured OTLP destination without explicit logging by the application. The issue affects MLflow 3.16.0 with CrewAI 1.15.18 and is fixed in MLflow 3.17.0.

Impact

Confidential CrewAI LLM provider API keys are exposed in MLflow trace telemetry data. This leakage can lead to unauthorized access to the LLM provider services if the telemetry data is accessed by unauthorized parties. The vulnerability does not affect integrity or availability but results in a confidentiality breach of sensitive credentials.

Summary

The vulnerability (CVE-2026-116691) occurs in MLflow's CrewAI autologging integration, specifically in the _get_agent_attributes() and _construct_full_inputs() functions. These functions serialize the CrewAI Agent object's __dict__, converting fields to strings without sanitizing credential-bearing objects such as the LLM instance. The LLM object's string representation includes the api_key in plaintext, which MLflow then records as span attributes and span inputs in trace telemetry. This results in unintentional leakage of API keys to MLflow's trace backend or any OTLP-compatible telemetry collector. The flaw was verified on MLflow 3.16.0 and CrewAI 1.15.18. The issue is fixed in MLflow version 3.17.0 by presumably improving sanitization and avoiding recursive serialization of sensitive fields.

CWE-200Exposure of Sensitive Information to an Unauthorized Actor (opens in a new tab)
Class weaknessLikelihood of exploit: HighCWE Top 25 (2025): #20OWASP Top 10 A01:2025: Broken Access Control (opens in a new tab)
The product exposes sensitive information to an actor that is not explicitly authorized to have access to that information.
MITRE discourages mapping vulnerabilities to this broad entry; the root cause is usually a more specific weakness.
Architecture and Design · Separation of Privilege
Compartmentalize the system to have "safe" areas where trust boundaries can be unambiguously drawn. Do not allow sensitive data to go outside of the trust boundary and always be careful when interfacing with a compartment outside of the safe area.
OWASP Cheat Sheets
Error Handling (opens in a new tab)
Logging (opens in a new tab)
Authorization (opens in a new tab)
OWASP ASVS 4.0.3
V8.3.4 (opens in a new tab)
V14.3.3 (opens in a new tab)
V7.3.3 (opens in a new tab)
Attack patterns
Subverting Environment Variable Values (opens in a new tab)
ICMP Echo Request Ping (opens in a new tab)
TCP SYN Scan (opens in a new tab)
Enumerate Mail Exchange (MX) Records (opens in a new tab)
DNS Zone Transfers (opens in a new tab)
From MITRE CWE, MITRE CAPEC and OWASP (Top 10, OpenCRE).
CWE-312Cleartext Storage of Sensitive Information (opens in a new tab)
Base weaknessOWASP Top 10 A06:2025: Insecure Design (opens in a new tab)
The product stores sensitive information in cleartext within a resource that might be accessible to another control sphere.
Implementation · System Configuration · Operation
When storing data in the cloud (e.g., S3 buckets, Azure blobs, Google Cloud Storage, etc.), use the provider's controls to encrypt the data at rest.
Implementation · System Configuration · Operation
In some systems/environments such as cloud, the use of "double encryption" (at both the software and hardware layer) might be required, and the developer might be solely responsible for both layers, instead of shared responsibility with the administrator of the broader system/environment.
OWASP Cheat Sheets
Secure Design Principles (opens in a new tab)
OWASP ASVS 4.0.3
V8.1.6 (opens in a new tab)
V8.2.3 (opens in a new tab)
V8.2.2 (opens in a new tab)
Attack patterns
Retrieve Embedded Sensitive Data (opens in a new tab)
From MITRE CWE, MITRE CAPEC and OWASP (Top 10, OpenCRE).
CWE-532Insertion of Sensitive Information into Log File (opens in a new tab)
Base weaknessLikelihood of exploit: MediumOWASP Top 10 A09:2025: Security Logging & Alerting Failures (opens in a new tab)
The product writes sensitive information to a log file.
Architecture and Design · Implementation
Consider seriously the sensitivity of the information written into log files. Do not write secrets into the log files.
Distribution
Remove debug log files before deploying the application into production.
Operation
Protect log files against unauthorized read/write.
OWASP Cheat Sheets
Application Logging Vocabulary (opens in a new tab)
Logging (opens in a new tab)
OWASP ASVS 4.0.3
V7.1.1 (opens in a new tab)
V7.1.2 (opens in a new tab)
V8.3.5 (opens in a new tab)
Attack patterns
Fuzzing for application mapping (opens in a new tab)
From MITRE CWE, MITRE CAPEC and OWASP (Top 10, OpenCRE).

References

Community reviews

Loading reviews…

Exploitation evidence

What to do

Affected software

Description

AI analysisMachine-generated, updated 27 Oct 2026, 22:27 UTC

WeaknessesCWE-200, CWE-312, CWE-532

Technical details

References

Community reviews

Related threats