Attack ChainAIDerived
Hypothetical attack flow derived from CVE metadata – not a confirmed exploit chain
-
ReconReach ComfyUI HTTP API over network
-
DeliverySubmit crafted POST /prompt workflow
-
ExploitLoadImage node bypasses input-dir validation
-
InstallPath traversal joins arbitrary host path
-
C2Server loads out-of-scope image file
-
ExecuteRetrieve file bytes via GET /view
-
ImpactExfiltrate sensitive host files
Vulnerability Assessment
| Weakness |
CWE
OWASP Top 10:2025
A01:2025 - Broken Access Control
ASVS 5.0
V5 - File Handling Path canonicalization, traversal prevention
MITRE ATT&CK
2 techniques across 2 tactics: Discovery, Collection The technique list with ATT&CK links and the kill-chain view is part of the Basic plan and up. |
| Exploitation Conditions | Exploitation requires network reachability to the ComfyUI HTTP API and the ability to submit a workflow via POST /prompt using one of the loader nodes (LoadImage, LoadImageMask, LoadImageOutput, LoadAudio, LoadLatent, LoadVideo, Load3D) with an attacker-controlled annotated filename; no authentication is needed (PR:N) and no user interaction is required (UI:N) against default configurations prior to 0.28.0. The specific enabling condition is that LoadImage's VALIDATE_INPUTS method makes the execution engine skip COMBO input-directory validation, and folder_paths performs no containment check. Limiting factors: successful exfiltration through /view is constrained to files the server can serve as image formats, so non-image file contents are not directly readable; deployments that require authentication in front of the API, restrict /prompt and /view, or are bound to localhost/isolated networks are not remotely exploitable; and already patched instances (>=0.28.0) are not affected. |
| Impact | The CVSS 3.1 vector CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:N/A:N describes a low-complexity, network-reachable, unauthenticated attack with high confidentiality impact and no integrity or availability impact - internally consistent with the description. Real risk is elevated because ComfyUI is frequently run without authentication and increasingly exposed to networks or the internet, and the exploit is a simple crafted POST /prompt requiring no user interaction. The offset is that impact is bounded: exfiltration through /view is limited to files the server can render as image formats, so it is a targeted file-disclosure / path-probing primitive rather than full arbitrary file read of any content, and it grants no code execution or write access. No EPSS score, KEV listing, SSVC decision, or POC was provided; absence of KEV/EPSS data means widespread automated exploitation cannot be confirmed. Overall this is a genuine, easily-triggered information-disclosure issue worth prompt patching, but the confidentiality-only, image-format-limited nature keeps it below RCE-class urgency. |
| Exploit ScenarioDerived | An attacker who can reach a ComfyUI instance's HTTP API sends a crafted POST /prompt containing a workflow whose LoadImage (or LoadImageOutput/Load3D) node supplies an annotated filename with ../ traversal sequences pointing outside the input directory. Because VALIDATE_INPUTS causes the engine to skip input-directory validation and no containment check is applied, the server resolves and loads the out-of-scope path, and the attacker then retrieves the rendered image bytes via GET /view to exfiltrate or confirm the existence of arbitrary host files. Given AV:N/AC:L/PR:N/UI:N, this requires only network reachability and no credentials, though no public PoC was provided in the input. |
Get an e-mail when this CVE is exploited, patched or re-scored. Free account.
Alert meAnalysisAIv1 – Jul 31, 2026
Path traversal arbitrary file read in ComfyUI (comfy-org) before 0.28.0 lets remote unauthenticated attackers probe arbitrary host filesystem paths and exfiltrate image-format files via the /view endpoint. A crafted POST /prompt workflow using LoadImage or sibling loader nodes joins attacker-controlled annotated filenames to a base directory with no containment check, and LoadImage's VALIDATE_INPUTS method causes the engine to skip input-directory validation. Rated CVSS 7.5 (confidentiality-only); no public exploit identified at time of analysis and it is not on CISA KEV.
RemediationAI
Vendor-released patch: upgrade to ComfyUI 0.28.0 or later, which adds the folder_paths.is_within_directory() containment check per advisory GHSA-rvxv-29p8-pxgq (https://github.com/Comfy-Org/ComfyUI/security/advisories/GHSA-rvxv-29p8-pxgq) and PR #14734 (https://github.com/Comfy-Org/ComfyUI/pull/14734). If you cannot upgrade immediately, do not expose the ComfyUI API to untrusted networks: bind it to localhost or place it behind a reverse proxy or VPN that enforces authentication, and restrict or block unauthenticated access to the POST /prompt and GET /view endpoints, since these are the ingestion and exfiltration paths - the trade-off is that remote and multi-user workflow submission will be blocked. Run the ComfyUI process under a low-privilege account with a restricted filesystem view (for example a container or chroot) so that traversal reads reach as few sensitive files as possible, accepting the operational overhead of containerization. Avoid loading untrusted workflow JSON from third parties until patched.
Affected VersionsENISA
Affected Packages
Source: OSV (Open Source Vulnerabilities)
Affects your stack
Affects 1 package in euvd
Checking this CVE against your own stacks and servers is part of the Basic plan and up.
Technical ContextAI · show details
ComfyUI is a modular node/graph GUI and API backend for diffusion (Stable Diffusion) image generation, exposing an HTTP API where workflows are submitted as JSON. The root cause is CWE-22 (Improper Limitation of a Pathname to a Restricted Directory / path traversal): folder_paths.get_annotated_filepath and exists_annotated_filepath build a filesystem path by joining a workflow-supplied 'annotated' filename (which encodes a base directory such as input/output/temp) to a base directory without verifying the result stays inside that directory, so sequences like ../../ escape it. Compounding this, the LoadImage node declares a VALIDATE_INPUTS method, which signals the execution engine to bypass the normal COMBO validation that would otherwise restrict the value to entries actually present in the input directory. The fix in PR #14734 introduces a reusable folder_paths.is_within_directory() containment helper that realpaths both the base directory and target (also defeating symlink escape) and requires commonpath to equal the base directory; the same PR hardens the model-preview and userdata endpoints and adds is_dangerous_content_type() to force risky MIME types to download.
Know when Comfyui gets its next vulnerability
- Watch products and get an email when something new is published
- Check this CVE against your own stack and servers
- Exploit and PoC references, threat intelligence from 22 sources
No credit card · 7-day full trial
Severity by source
AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:N/A:N
Unauthenticated network POST /prompt with no user interaction and low complexity yields confidentiality-only file disclosure; no integrity or availability impact, so I:N/A:N.
Primary rating from Vendor (GitHub_M).
CVSS VectorVendor: GitHub_M
Lifecycle Timeline
2DescriptionCVE.org
ComfyUI is a modular diffusion model GUI, API, and backend with a graph-and-node interface. Prior to 0.28.0, folder_paths.get_annotated_filepath and exists_annotated_filepath join workflow-controlled annotated filenames to a base directory without a containment check, allowing an unauthenticated crafted POST /prompt workflow using LoadImage or sibling nodes to probe arbitrary host paths and exfiltrate image-format files through /view. LoadImage defines a VALIDATE_INPUTS method, which causes the execution engine to skip COMBO (input-directory) validation. Affected nodes include LoadImage, LoadImageMask, LoadImageOutput, LoadAudio, LoadLatent, LoadVideo, and Load3D. This issue is fixed in version 0.28.0.
Executive ViewAI
ComfyUI versions before 0.28.0 contain a path traversal vulnerability enabling remote unauthenticated attackers to exfiltrate arbitrary files from the host system, presenting immediate confidentiality risk to training data, model artifacts, and sensitive image files stored on affected instances. With no vendor patch currently available, organizations must implement compensating controls urgently. No public exploit has been identified at time of analysis, reducing immediate exploitation risk but not eliminating the underlying vulnerability.
No vendor-released patch identified at time of analysis. Interim mitigations: (1) Restrict network access to ComfyUI instances to authorized users and networks via firewall or VPN; (2) Disable or isolate the /view endpoint if not operationally critical; (3) Implement Web Application Firewall rules blocking path traversal patterns (e.g., '../', '..\') and monitor POST /prompt request anomalies; (4) Enable detailed access logging on API endpoints and audit for exploitation patterns; (5) Disable LoadImage and sibling loader nodes if alternative workflows exist; (6) Relocate sensitive image files and training data outside directories accessible to ComfyUI instances; (7) Subscribe to ComfyUI security advisories and vendor releases for patch notification. Deploy full patch to all instances within 24 hours of vendor release.
Threat Intelligence Overview
How this decision was reached
The engine reads each signal below and applies a fixed list of rules from top to bottom. The first rule that matches sets the recommended action – like a statement, every line is shown.
What every action level means
SSVC AssessmentENISA
Priority Score
Attack Technique
Path traversal exploits occur when applications use user-controlled input to construct file system paths without proper validation.
Arbitrary file write via path traversal in ComfyUI before 0.30.0 lets an attacker place attacker-controlled content anyw
comfyanonymous/comfyui version v0.2.4 suffers from a non-blind Server-Side Request Forgery (SSRF) vulnerability. Rated h
A stored cross-site scripting (XSS) vulnerability exists in comfyanonymous/comfyui version 0.2.2 and possibly earlier. R
Remote code execution in ComfyUI v0.23.0 (and all prior 0.x releases) lets unauthenticated attackers run arbitrary Pytho
Stored cross-site scripting in ComfyUI (Comfy-Org) before 0.28.0 lets an attacker upload HTML or SVG to the per-user use
Stored cross-site scripting in ComfyUI before 0.28.0 lets an attacker upload a malicious SVG that the /view endpoint ser
Unauthenticated information disclosure in ComfyUI (comfy-org/comfyui) before 0.28.0 lets remote attackers read arbitrary
Path traversal in ComfyUI up to version 0.13.0 allows authenticated remote attackers to read arbitrary files on the serv
Path traversal in ComfyUI up to version 0.13.0 allows authenticated remote attackers to read arbitrary files via manipul
Cross-site request forgery (CSRF) in ComfyUI up to version 0.13.0 allows unauthenticated remote attackers to modify appl
Cross-site scripting (XSS) in ComfyUI up to version 0.13.0 allows authenticated remote attackers to inject malicious scr
Stored cross-site scripting (XSS) in ComfyUI's userdata endpoint (getuserdata function in app/user_manager.py) allows au
EUVD-2026-51403
NVD
CVE.org
EPSS
Vulnrichment
GitHub
VulDB
Exploit-DB