CVE-2026-104478: Formwork: path traversal in backups reads/deletes files
HIGHFormwork, a flat-file PHP CMS, has a path traversal flaw in its BackupController before 2.3.13. An authenticated panel user with backup download or delete permission can send a base64-encoded, backslash-separated payload that slips past PHP basename on Linux and reads or deletes arbitrary files outside the backup directory. Exposure is moderate: the CVSS is 7.1 and the attack needs low privileges and no user interaction. The EPSS score of 0.00352 (73rd percentile) is low, there is no public exploit or Nuclei template, and the CVE is not in CISA KEV. Its link to AI is weak, since Formwork is a generic CMS and not an ML component. Upgrade to 2.3.13 or later. Until then, review who holds backup permissions and check any Formwork instance that hosts AI documentation, model cards, or portals for exposed secrets.
What is the risk?
Medium-high on a standalone basis (CVSS 7.1: high confidentiality impact, low availability impact, no integrity impact). The attacker needs a valid panel account with backup download or delete permission, which limits exposure. Exploitation likelihood is low (EPSS 0.35%, no public exploit, not in KEV). The AI/ML relevance is indirect: the risk only matters if a Formwork instance sits next to AI assets, such as a documentation or model-portal site on a server that holds API keys or configuration files.
How does the attack unfold?
What systems are affected?
| Package | Ecosystem | Vulnerable Range | Patched |
|---|---|---|---|
| formwork | — | — | No patch |
Do you use formwork? You're affected.
How severe is it?
What is the attack surface?
What should I do?
1 step1) Upgrade Formwork to 2.3.13 or later (patch commit 89e7821). 2) Review which panel users hold backup download or delete permissions and remove any that are not needed. 3) Run the web server with least privilege and separate it from secrets, using environment variables or a secrets manager outside the web root. 4) Check web and panel logs for backup requests with base64 parameters that decode to backslashes or '..' sequences. 5) Check for unexpected file deletions or reads outside the backups directory. 6) Isolate the CMS from hosts that run AI workloads.
What does CISA's SSVC say?
Source: CISA Vulnrichment (SSVC v2.0). Decision based on the CISA Coordinator decision tree.
How is it classified?
Which compliance frameworks are affected?
This CVE is relevant to:
Frequently Asked Questions
What is CVE-2026-104478?
Formwork, a flat-file PHP CMS, has a path traversal flaw in its BackupController before 2.3.13. An authenticated panel user with backup download or delete permission can send a base64-encoded, backslash-separated payload that slips past PHP basename on Linux and reads or deletes arbitrary files outside the backup directory. Exposure is moderate: the CVSS is 7.1 and the attack needs low privileges and no user interaction. The EPSS score of 0.00352 (73rd percentile) is low, there is no public exploit or Nuclei template, and the CVE is not in CISA KEV. Its link to AI is weak, since Formwork is a generic CMS and not an ML component. Upgrade to 2.3.13 or later. Until then, review who holds backup permissions and check any Formwork instance that hosts AI documentation, model cards, or portals for exposed secrets.
Is CVE-2026-104478 actively exploited?
No confirmed active exploitation of CVE-2026-104478 has been reported, but organizations should still patch proactively.
How to fix CVE-2026-104478?
1) Upgrade Formwork to 2.3.13 or later (patch commit 89e7821). 2) Review which panel users hold backup download or delete permissions and remove any that are not needed. 3) Run the web server with least privilege and separate it from secrets, using environment variables or a secrets manager outside the web root. 4) Check web and panel logs for backup requests with base64 parameters that decode to backslashes or '..' sequences. 5) Check for unexpected file deletions or reads outside the backups directory. 6) Isolate the CMS from hosts that run AI workloads.
What systems are affected by CVE-2026-104478?
This vulnerability affects the following AI/ML architecture patterns: Web portals and documentation sites for AI products, Self-hosted CMS co-located with AI services, Content management front-ends for ML platforms.
What is the CVSS score for CVE-2026-104478?
CVE-2026-104478 has a CVSS v3.1 base score of 7.1 (HIGH). The EPSS exploitation probability is 0.35%.
What is the AI security impact?
Affected AI Architectures
MITRE ATLAS Techniques
AML.T0012Valid AccountsAML.T0037Data from Local SystemAML.T0049Exploit Public-Facing ApplicationCompliance Controls Affected
What are the technical details?
Original Advisory
Formwork before 2.3.13 contains a path traversal vulnerability in BackupController that allows authenticated panel users to read or delete arbitrary files. Attackers with backup download or delete permission can supply a base64-encoded backslash-separated traversal payload that bypasses PHP basename on Linux to access files outside the backup directory.
Exploitation Scenario
An attacker obtains low-privilege panel credentials, for example through phishing or a reused password, for a Formwork site that hosts an AI product's documentation. The account has backup download permission. The attacker sends a request to the backup download endpoint with a base64-encoded path using backslash separators, such as '..\..\.env'. On Linux, PHP basename does not treat the backslash as a separator, so the check passes and the file is served from outside the backup directory. The attacker obtains LLM provider API keys and database credentials, then uses them against the organization's AI services. Using the delete permission, the attacker could also remove files to disrupt the site.
Weaknesses (CWE)
CWE-22Improper Limitation of a Pathname to a Restricted DirectoryPrimaryCWE-22Improper Limitation of a Pathname to a Restricted DirectoryCWE-22 — Improper Limitation of a Pathname to a Restricted Directory: 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.
CVSS Vector
CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:H/I:N/A:LReferences
Timeline
Related Vulnerabilities
CVE-2025-512010.0smolagents: sandbox escape enables unauthenticated RCE
Same attack type: Data LeakageCVE-2023-376510.0MLflow: path traversal allows arbitrary file read
Same attack type: Data LeakageCVE-2025-282810.0LangChain RequestsToolkit: SSRF exposes cloud metadata
Same attack type: Auth BypassCVE-2025-5376710.0Azure OpenAI: SSRF EoP, no auth required (CVSS 10)
Same attack type: Auth BypassCVE-2026-2185810.0n8n: Input Validation flaw enables exploitation
Same attack type: Auth Bypass