CVE-2026-69088: Grav CMS: callable guard bypass enables file read/write

GHSA-7pgq-cr25-xvc8HIGHPoC AVAILABLECISA: ATTEND
Published September 17, 2026
CISO Take

Grav CMS (getgrav/grav 2.0.7–2.0.10) has a flaw in its blueprint dynamic-field guard: any callable written as a fully-qualified Class::method string skips the dangerous-function denylist. A low-privileged page editor (admin.pages, not super-admin) can therefore invoke arbitrary public static PHP methods and read any server-readable file, including account YAML files with admin bcrypt hashes, which are served to anonymous visitors of the crafted page. The same path allows creating or copying files and directories as the web-server account. Exploitation likelihood is modest today: EPSS is 0.4% (top 67%), no public exploit or Nuclei template exists, and it is not in CISA KEV. But the attack needs only a low-privilege editor account and a single page save, and the package has 504 downstream dependents and 76 prior CVEs. Upgrade to 2.0.11, review who holds admin.pages, and audit page frontmatter for data-*@ directives containing '::'. Grav is a general-purpose CMS, so this matters mainly where it hosts AI/ML documentation, model cards or dashboards.

Sources:NVDGitHub AdvisoryEPSSOpenSSFATLASvulncheck.com

What is the risk?

High (CVSS 8.1). Network-exploitable with low complexity, but it needs an authenticated account with page-editing rights, which limits the attacker pool. Confidentiality and integrity impact are both high: arbitrary file read and arbitrary file or directory creation as the web-server user. Exposure is limited to Grav 2.0.7–2.0.10 with the admin and form plugins. The AI/ML link is indirect, since Grav is a CMS and not an AI framework. Risk rises where compromised credentials or config files give access to AI platforms. There is no known active exploitation, and the EPSS score is low.

How does the attack unfold?

Initial Access
Attacker logs in to Grav admin with a low-privilege admin.pages editor account.
AML.T0012
Exploitation
Attacker saves page frontmatter with a data-opts@ directive using a Class::method callable that bypasses the guard.
AML.T0049
Collection
Anonymous request to the crafted page runs Utils::download and returns files such as account YAML with password hashes.
AML.T0037
Impact
Attacker cracks hashes or uses leaked secrets to escalate, and uses Folder::copy, move or delete to tamper with site content.

What systems are affected?

PackageEcosystemVulnerable RangePatched
Panelcomposer>= 2.0.7, <= 2.0.102.0.11
5.8KOpenSSF 6.5514 dependentsPushed 3d ago67% patched~15d to patchFull package profile →

Do you use Panel? You're affected.

How severe is it?

CVSS 3.1
8.1 / 10
EPSS
0.3%
chance of exploitation in 30 days
Higher than 22% of all CVEs
Exploitation Status
Exploit Available
Exploitation: MEDIUM
Sophistication
Trivial
Exploitation Confidence
medium
○CISA SSVC: Public PoC
○Public PoC indexed (trickest/cve)
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?

AVACPRUISCIA
AVNetwork
ACLow
PRLow
UINone
SUnchanged
CHigh
IHigh
ANone

What should I do?

1 step
  1. 1) Upgrade getgrav/grav to 2.0.11 or later. 2) Until patched, revoke or restrict admin.pages from untrusted users and enforce MFA on admin accounts. 3) Search user/pages frontmatter for data-*@ directives containing '::' and review any unexpected ones. 4) Rotate administrator passwords and any secrets stored on the server (API keys, .env contents) if abuse is suspected. 5) Monitor web logs for POSTs to /admin/pages/* with form-field frontmatter, and for unexpected new page directories under user/pages/. 6) Run the web server with minimal filesystem permissions and keep secrets outside the web root. 7) Consider a WAF rule to flag 'Class::method' strings in admin page-save requests.

What does CISA's SSVC say?

DecisionAttend
Exploitationpoc
AutomatableNo
Technical Impacttotal

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:

ISO 42001
A.7.4-Quality of data for AI systems
NIST AI RMF
MANAGE 2.3-Responses to unknown risks and incident management
OWASP LLM Top 10
LLM03-Supply Chain

Frequently Asked Questions

What is CVE-2026-69088?

Grav CMS (getgrav/grav 2.0.7–2.0.10) has a flaw in its blueprint dynamic-field guard: any callable written as a fully-qualified Class::method string skips the dangerous-function denylist. A low-privileged page editor (admin.pages, not super-admin) can therefore invoke arbitrary public static PHP methods and read any server-readable file, including account YAML files with admin bcrypt hashes, which are served to anonymous visitors of the crafted page. The same path allows creating or copying files and directories as the web-server account. Exploitation likelihood is modest today: EPSS is 0.4% (top 67%), no public exploit or Nuclei template exists, and it is not in CISA KEV. But the attack needs only a low-privilege editor account and a single page save, and the package has 504 downstream dependents and 76 prior CVEs. Upgrade to 2.0.11, review who holds admin.pages, and audit page frontmatter for data-*@ directives containing '::'. Grav is a general-purpose CMS, so this matters mainly where it hosts AI/ML documentation, model cards or dashboards.

Is CVE-2026-69088 actively exploited?

Proof-of-concept exploit code is publicly available for CVE-2026-69088, increasing the risk of exploitation.

How to fix CVE-2026-69088?

1) Upgrade getgrav/grav to 2.0.11 or later. 2) Until patched, revoke or restrict admin.pages from untrusted users and enforce MFA on admin accounts. 3) Search user/pages frontmatter for data-*@ directives containing '::' and review any unexpected ones. 4) Rotate administrator passwords and any secrets stored on the server (API keys, .env contents) if abuse is suspected. 5) Monitor web logs for POSTs to /admin/pages/* with form-field frontmatter, and for unexpected new page directories under user/pages/. 6) Run the web server with minimal filesystem permissions and keep secrets outside the web root. 7) Consider a WAF rule to flag 'Class::method' strings in admin page-save requests.

What systems are affected by CVE-2026-69088?

This vulnerability affects the following AI/ML architecture patterns: Web CMS front ends for AI products, ML documentation and portal sites, Internal AI knowledge bases.

What is the CVSS score for CVE-2026-69088?

CVE-2026-69088 has a CVSS v3.1 base score of 8.1 (HIGH). The EPSS exploitation probability is 0.29%.

What is the AI security impact?

Affected AI Architectures

Web CMS front ends for AI productsML documentation and portal sitesInternal AI knowledge bases

MITRE ATLAS Techniques

AML.T0037Data from Local System
AML.T0049Exploit Public-Facing Application
AML.T0055Unsecured Credentials

Compliance Controls Affected

ISO 42001: A.7.4
NIST AI RMF: MANAGE 2.3
OWASP LLM Top 10: LLM03

What are the technical details?

Original Advisory

### Summary Grav CMS's blueprint dynamic-field callable guard can be bypassed with a fully-qualified `Class::method` string, letting an account with only page-editing rights (`admin.pages`, not super-admin) plant a directive in a page's form-field frontmatter that invokes an arbitrary public static PHP method with attacker-controlled arguments. Using built-in gadget methods this yields, at minimum, arbitrary reading of any server-readable file (disclosed to anonymous visitors of the crafted page) and arbitrary creation/copying of files and directories under the web-server account. ### Details `Blueprint::isSafeDynamicCall()` (`system/src/Grav/Common/Data/Blueprint.php`, method around line 488) is meant to block dangerous callables named in a blueprint's dynamic-field directives (`data-*@`). It only consults its dangerous-name denylist when the callable string does **not** contain `::`: ```php if (is_string($function) && !str_contains($function, '::') && Utils::isDangerousFunction($function)) { return false; } ``` Any callable string containing `::` — i.e. every `Class::method` static call — skips the check entirely and is passed to `call_user_func_array()` at `Blueprint::dynamicData()` (`Blueprint.php`, around line 461) and `FlexDirectory::dynamicDataField()` (`system/src/Grav/Framework/Flex/FlexDirectory.php`, around line 937). There is no allowlist restricting which classes or methods may be invoked this way; only the call's *arguments* are (separately) scanned for smuggled dangerous callables, never the target itself. `Utils::isDangerousFunction()` (`system/src/Grav/Common/Utils.php`) classifies any string containing a colon (`str_contains($name, ":")`) or a namespace backslash as dangerous — so a qualified `Class::method` string would be rejected *if* it ever reached this function. The `!str_contains($function, '::')` condition in `isSafeDynamicCall()` ensures it never does, which is what leaves qualified static calls entirely unscreened. (Whether the exemption was intended to admit legitimate `Class::method` option-providers is a plausible reading of the surrounding code, but the intent is not established here.) This is an incomplete fix of two recently published advisories — one addressing page editors executing hidden callables via form-field settings, the other extending the same guard to Flex directories. The guard those fixes introduced never covered qualified static calls. Grav's own permission model separates page-content code execution into a distinct, higher privilege (`admin.pages_twig`) from plain page editing (`admin.pages`), so invoking arbitrary methods from an `admin.pages`-authored page is a genuine trust-boundary bypass, not editor capability by design. ### PoC Tested on Grav `develop` at commit `db8c1fc` (which self-reports version 2.0.11) with the admin and form plugins, using an account granted only `admin.login` + `admin.pages` (page editor, not super-admin). The same guard is present in every current release from 2.0.7 through 2.0.10. Base URL shown as `https://grav.example`. **A. Arbitrary file read (confidentiality)** 1. Log in to `/admin` as the page-editor account (GET `/admin` for the login nonce, then POST `task=login`). 2. Save a page via the standard admin endpoint, `POST /admin/pages/<route>` with the session cookie, the admin nonce, and `data[frontmatter]` containing a form field with a `download` gadget directive: ```yaml forms: x: fields: y: type: text data-opts@: - 'Grav\Common\Utils::download' - '/etc/passwd' - false - 0 - 1024 - mime: 'text/plain' ``` The server returns `HTTP 200` and accepts the save — the `data-opts@` directive is not rejected. 3. As an unauthenticated visitor (no cookies), request the saved page, e.g. `GET /<route>` (use a fresh query string to avoid a cached copy; immediately after saving, a first request may `404` while the flat-file page index catches up — retry moments later). The response is `HTTP 200` with the raw contents of `/etc/passwd` in the body (`root:x:0:0:...`). Pointing the path at `user/accounts/<name>.yaml` instead returns that account file, including its `hashed_password:` bcrypt line — i.e. an anonymous visitor obtains a stored administrator's password hash. **B. Arbitrary file/directory write (integrity) — verified** Using the same mechanism with `Grav\Common\Filesystem\Folder::copy` (a public static method taking source and destination paths), a page editor caused the server to copy an existing page directory to an attacker-chosen new path under `user/pages/`; the newly created page then rendered its (attacker-controlled) content at the new route over plain HTTP. This demonstrates attacker-controlled creation of files/directories anywhere the web-server account can write. `Folder::move` and `Folder::delete` are equally reachable (their destructive nature was not exercised). ### Impact A page editor (an `admin.pages`-only account, **not** super-admin) can, through a page they author: - **Read any server-readable file**, disclosed to any anonymous, unauthenticated visitor of the crafted page — including `user/accounts/*.yaml`, which stores account metadata and bcrypt password hashes. An attacker may attempt offline cracking of a disclosed hash; recovery of a weak or reused administrator password could lead to full admin-panel compromise. Other secrets on disk (site/plugin config, environment files) are equally exposed. - **Create or overwrite files and directories** under the web-server account (demonstrated via `Folder::copy`), with `Folder::move`/`Folder::delete` additionally reachable for destructive tampering. Because the guard permits *any* public static method, the reachable impact is bounded only by the gadget surface of the loaded codebase, not by this report's demonstrated cases.

Exploitation Scenario

An attacker obtains or already holds a page-editor account on a Grav site hosting an AI product's documentation portal. They log in to /admin and save a page whose frontmatter contains a form field with data-opts@ set to Grav\Common\Utils::download with /etc/passwd or user/accounts/admin.yaml as the path. An anonymous request to the crafted page returns the file contents, including the admin bcrypt hash. The attacker cracks a weak hash offline, escalates to super-admin and reads config files holding LLM API keys or cloud credentials. They can also use Folder::copy to plant attacker-controlled pages or content on the site.

Weaknesses (CWE)

CWE-200 — Exposure of Sensitive Information to an Unauthorized Actor: The product exposes sensitive information to an actor that is not explicitly authorized to have access to that information.

  • [Architecture and Design] 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. Ensure that appropriate compartmentalization is built into the system design, and the compartmentalization allows for and reinforces privilege separation functionality. Architects and designers should rely on the principle of least privilege to decide the appropriate time to use privileges and the time to drop privileges.

Source: MITRE CWE corpus.

CVSS Vector

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

Timeline

Published
September 17, 2026
Last Modified
October 8, 2026
First Seen
September 17, 2026

Related Vulnerabilities