CVE-2026-32981: Ray unauthenticated path traversal

Q: Is CVE-2026-32981 actively exploited?

No confirmed active exploitation of CVE-2026-32981 has been reported, but organizations should still patch proactively.

Q: How to fix CVE-2026-32981?

1. Patch: Upgrade ray to version 2.8.1 or later immediately — this is the only complete fix. 2. Immediate network workaround: Bind the Ray Dashboard to 127.0.0.1 only (`ray start --dashboard-host=127.0.0.1`) or firewall port 8265 to localhost or a trusted management VLAN. Ray Dashboard should never be internet-exposed. 3. Detection: Search web server and proxy logs for requests containing '../', '%2e%2e%2f', '%2e%2e/', or '..%2f' on port 8265. Alert on any static file request resolving outside the Ray static assets directory. 4. Cloud audit: Review CloudTrail, GCP Audit Logs, or Azure Monitor for credential usage from unexpected source IPs or regions if Ray instances were internet-accessible. 5. Credential rotation: If exposure is suspected, immediately rotate cloud access keys, API tokens, and database passwords present on affected hosts. 6. Inventory: Use your asset inventory or 'ray status' to identify all Ray cluster head nodes and verify dashboard binding.

Q: What systems are affected by CVE-2026-32981?

This vulnerability affects the following AI/ML architecture patterns: distributed training pipelines, model serving infrastructure, ML Ops platforms, reinforcement learning environments, Ray-based agent frameworks.

Q: What is the CVSS score for CVE-2026-32981?

CVE-2026-32981 has a CVSS v3.1 base score of 7.5 (HIGH). The EPSS exploitation probability is 0.93%.

CISO Take

Ray's distributed ML platform dashboard (port 8265, enabled by default) contains an unauthenticated path traversal vulnerability that allows any network-reachable attacker to read arbitrary files from the host without credentials or user interaction. With 872 downstream dependents and Ray widely deployed in ML training clusters, Ray Serve inference environments, and distributed data pipelines, the blast radius spans a significant portion of production MLOps infrastructure. The CVSS metrics (AV:N/AC:L/PR:N/UI:N) place this firmly in the automated-scanner-reachable category, and a PacketStorm reference suggests public exploitation interest even without a formal Nuclei template. Patch to ray 2.8.1 immediately; if patching is delayed, firewall port 8265 to localhost or a trusted management VLAN and audit access logs for traversal sequences.

Sources: NVD EPSS GitHub Advisory ATLAS OpenSSF PacketStorm VulnCheck

What is the risk?

High risk for any organization operating Ray-based MLOps infrastructure. The attack requires no authentication, no special privileges, and no user interaction — trivially exploitable by automated scanners. Ray dashboards are routinely deployed on internal ML cluster nodes with broad network reachability and absent authentication controls, particularly in cloud-hosted training environments where default security group configurations may expose port 8265. The 9 prior CVEs in the same package and an OpenSSF Scorecard of 5.7/10 indicate persistent security debt. Successful exploitation exposes host filesystem contents including cloud provider credentials, API keys, environment variable files, and proprietary model configurations, enabling multi-stage attacks beyond the initial file disclosure.

How does the attack unfold?

Reconnaissance

Attacker scans for exposed Ray Dashboard instances on port 8265 using Shodan, Censys, or targeted port scans against cloud CIDR ranges, identifying unauthenticated ML cluster management interfaces.

AML.T0006

Initial Access

Attacker sends an unauthenticated HTTP GET request to the Ray Dashboard static file handler with path traversal sequences (e.g., /static/../../../../etc/passwd), bypassing directory restrictions.

AML.T0049

Data Collection

Attacker iterates traversal requests to exfiltrate high-value files: cloud credential files (~/.aws/credentials), .env configs, Kubernetes service account tokens, and environment variables containing API keys.

AML.T0037

Impact

Harvested credentials enable lateral movement into cloud infrastructure (S3, GCS, Azure Blob), access to proprietary training datasets and model artifacts, or further compromise of connected ML systems.

AML.T0055

Reconnaissance

Attacker scans for exposed Ray Dashboard instances on port 8265 using Shodan, Censys, or targeted port scans against cloud CIDR ranges, identifying unauthenticated ML cluster management interfaces.

AML.T0006

Initial Access

Attacker sends an unauthenticated HTTP GET request to the Ray Dashboard static file handler with path traversal sequences (e.g., /static/../../../../etc/passwd), bypassing directory restrictions.

AML.T0049

Data Collection

Attacker iterates traversal requests to exfiltrate high-value files: cloud credential files (~/.aws/credentials), .env configs, Kubernetes service account tokens, and environment variables containing API keys.

AML.T0037

Impact

Harvested credentials enable lateral movement into cloud infrastructure (S3, GCS, Azure Blob), access to proprietary training datasets and model artifacts, or further compromise of connected ML systems.

AML.T0055

What systems are affected?

Package	Ecosystem	Vulnerable Range	Patched
Ray	pip	< 2.8.1	`2.8.1`
43.4K OpenSSF 5.7 619 dependents Pushed 6d ago 81% patched ~112d to patch Full package profile →

Do you use Ray? You're affected.

How severe is it?

CVSS 3.1

7.5 / 10

EPSS

0.9%

chance of exploitation in 30 days

Higher than 57% of all CVEs

Source: EPSS v3 — FIRST.org

Exploitation Status

Exploit Available

Exploitation: MEDIUM

Sophistication

Trivial

Exploitation Confidence

medium

○ CISA SSVC: Public PoC

Composite signal derived from CISA KEV, VulnCheck KEV, CISA SSVC, EPSS, Metasploit, Exploit-DB, trickest/cve, Nuclei templates, and inthewild.io exploitation reports.

What is the attack surface?

AV Network

AC Low

PR None

UI None

S Unchanged

C High

I None

A None

What should I do?

7 steps

Patch: Upgrade ray to version 2.8.1 or later immediately — this is the only complete fix.
Immediate network workaround: Bind the Ray Dashboard to 127.0.0.1 only (ray start --dashboard-host=127.0.0.1) or firewall port 8265 to localhost or a trusted management VLAN. Ray Dashboard should never be internet-exposed.
Detection: Search web server and proxy logs for requests containing '../', '%2e%2e%2f', '%2e%2e/', or '..%2f' on port
Alert on any static file request resolving outside the Ray static assets directory.
Cloud audit: Review CloudTrail, GCP Audit Logs, or Azure Monitor for credential usage from unexpected source IPs or regions if Ray instances were internet-accessible.
Credential rotation: If exposure is suspected, immediately rotate cloud access keys, API tokens, and database passwords present on affected hosts.
Inventory: Use your asset inventory or 'ray status' to identify all Ray cluster head nodes and verify dashboard binding.

What does CISA's SSVC say?

Decision Track*

Exploitation poc

Automatable Yes

Technical Impact partial

Source: CISA Vulnrichment (SSVC v2.0). Decision based on the CISA Coordinator decision tree.

How is it classified?

Data Extraction Auth Bypass Framework Inference AML.T0006 - Active Scanning AML.T0037 - Data from Local System AML.T0049 - Exploit Public-Facing Application AML.T0055 - Unsecured Credentials

Which compliance frameworks are affected?

This CVE is relevant to:

EU AI Act

Article 15 - Accuracy, robustness and cybersecurity

ISO 42001

A.6.2 - AI system design and development

NIST AI RMF

MANAGE 2.2 - Mechanisms to sustain AI value and manage risk

OWASP LLM Top 10

LLM06 - Sensitive Information Disclosure

Frequently Asked Questions

What is CVE-2026-32981?

Ray's distributed ML platform dashboard (port 8265, enabled by default) contains an unauthenticated path traversal vulnerability that allows any network-reachable attacker to read arbitrary files from the host without credentials or user interaction. With 872 downstream dependents and Ray widely deployed in ML training clusters, Ray Serve inference environments, and distributed data pipelines, the blast radius spans a significant portion of production MLOps infrastructure. The CVSS metrics (AV:N/AC:L/PR:N/UI:N) place this firmly in the automated-scanner-reachable category, and a PacketStorm reference suggests public exploitation interest even without a formal Nuclei template. Patch to ray 2.8.1 immediately; if patching is delayed, firewall port 8265 to localhost or a trusted management VLAN and audit access logs for traversal sequences.

Is CVE-2026-32981 actively exploited?

No confirmed active exploitation of CVE-2026-32981 has been reported, but organizations should still patch proactively.

How to fix CVE-2026-32981?

1. Patch: Upgrade ray to version 2.8.1 or later immediately — this is the only complete fix. 2. Immediate network workaround: Bind the Ray Dashboard to 127.0.0.1 only (`ray start --dashboard-host=127.0.0.1`) or firewall port 8265 to localhost or a trusted management VLAN. Ray Dashboard should never be internet-exposed. 3. Detection: Search web server and proxy logs for requests containing '../', '%2e%2e%2f', '%2e%2e/', or '..%2f' on port 8265. Alert on any static file request resolving outside the Ray static assets directory. 4. Cloud audit: Review CloudTrail, GCP Audit Logs, or Azure Monitor for credential usage from unexpected source IPs or regions if Ray instances were internet-accessible. 5. Credential rotation: If exposure is suspected, immediately rotate cloud access keys, API tokens, and database passwords present on affected hosts. 6. Inventory: Use your asset inventory or 'ray status' to identify all Ray cluster head nodes and verify dashboard binding.

What systems are affected by CVE-2026-32981?

This vulnerability affects the following AI/ML architecture patterns: distributed training pipelines, model serving infrastructure, ML Ops platforms, reinforcement learning environments, Ray-based agent frameworks.

What is the CVSS score for CVE-2026-32981?

CVE-2026-32981 has a CVSS v3.1 base score of 7.5 (HIGH). The EPSS exploitation probability is 0.93%.

What is the AI security impact?

Affected AI Architectures

distributed training pipelinesmodel serving infrastructureML Ops platformsreinforcement learning environmentsRay-based agent frameworks

MITRE ATLAS Techniques

AML.T0006 Active Scanning

AML.T0037 Data from Local System

AML.T0049 Exploit Public-Facing Application

AML.T0055 Unsecured Credentials

Compliance Controls Affected

EU AI Act: Article 15

ISO 42001: A.6.2

NIST AI RMF: MANAGE 2.2

OWASP LLM Top 10: LLM06

What are the technical details?

Original Advisory

A path traversal vulnerability was identified in Ray Dashboard (default port 8265) in Ray versions prior to 2.8.1. Due to improper validation and sanitization of user-supplied paths in the static file handling mechanism, an attacker can use traversal sequences (e.g., ../) to access files outside the intended static directory, resulting in local file disclosure.

Exploitation Scenario

An attacker scans for exposed Ray Dashboard instances on port 8265 using Shodan, Censys, or a targeted port scan against known cloud CIDR ranges. Finding an unauthenticated dashboard, the attacker sends: GET /static/../../../../home/ubuntu/.aws/credentials HTTP/1.1. The Ray static file handler, failing to sanitize the traversal sequences, resolves the path and returns the AWS credentials file directly. The attacker pivots immediately to the AWS API using the harvested access key, enumerating S3 buckets — finding training datasets, model checkpoints, and potentially PII used in model fine-tuning. In a Kubernetes-hosted Ray deployment, the attacker reads /var/run/secrets/kubernetes.io/serviceaccount/token, gaining cluster API access and the ability to pivot to other workloads, exfiltrate further artifacts, or deploy a reverse shell via a manipulated pod spec.

Weaknesses (CWE)

CWE-22 Improper Limitation of a Pathname to a Restricted Directory ('Path Traversal') Primary CWE-22 Improper Limitation of a Pathname to a Restricted Directory ('Path Traversal')

CWE-22 — Improper Limitation of a Pathname to a Restricted Directory ('Path Traversal'): The product uses external input to construct a pathname that is intended to identify a file or directory that is located underneath a restricted parent directory, but the product does not properly neutralize special elements within the pathname that can cause the pathname to resolve to a location that is outside of the restricted directory.

[Implementation] Assume all input is malicious. Use an "accept known good" input validation strategy, i.e., use a list of acceptable inputs that strictly conform to specifications. Reject any input that does not strictly conform to specifications, or transform it into something that does. When performing input validation, consider all potentially relevant properties, including length, type of input, the full range of acceptable values, missing or extra inputs, syntax, consistency across related fields, and conformance to business rules. As an example of business rule logic, "boat" may be syntactically valid because it only contains alphanumeric characters, but it is not valid if the input is only expected to contain colors such as "red" or "blue." Do not rely exclusively on looking for malicious or malformed inputs. This is likely to miss at least one undesirable input, especially if the code's environment changes. This can give attackers enough room to bypass the intended validation. However, denylis
[Architecture and Design] For any security checks that are performed on the client side, ensure that these checks are duplicated on the server side, in order to avoid CWE-602. Attackers can bypass the client-side checks by modifying values after the checks have been performed, or by changing the client to remove the client-side checks entirely. Then, these modified values would be submitted to the server.

Source: MITRE CWE corpus.