Feb 20, 2026·6 min read·62 visits
API Platform's GraphQL implementation dropped the ball on the Relay `node` query. By asking for a resource via its global ID (IRI) through the `node` field, the system failed to load the specific security rules attached to that resource. This means an unauthenticated attacker could access data restricted to admins simply by knowing the resource's ID URL.
A critical authentication bypass vulnerability in API Platform Core's GraphQL subsystem allows attackers to read protected resources by leveraging the Relay `node` interface. The flaw stems from a failure to correctly load security operations when resolving Global Object Identifiers (IRIs), effectively neutralizing access control lists (ACLs) for specific queries.
GraphQL is magical. It lets front-end developers ask for exactly what they want, and the backend just... provides. One of the most powerful features in the GraphQL ecosystem is the Relay Global Object Identification specification. It introduces a standardized interface called Node. The promise is simple: give me a global ID (usually an IRI like /users/1), and I will return the object, no matter what type it is.
It’s the ultimate skeleton key for data fetching. You don't need to know if you are querying a User, a Product, or a SecretDocument; you just ask for the Node, and the server resolves the type for you. But here is the problem with magic skeleton keys: if you forget to check who is holding them, you are going to have a bad time.
In API Platform, a framework designed to automagically build APIs around your PHP classes, this convenience feature had a nasty side effect. While the framework is usually paranoid about checking @Security annotations and voters before handing out data, the node query path found a loophole. It was the digital equivalent of a bouncer checking IDs at the VIP entrance, while leaving the loading dock wide open because 'surely nobody knows where the loading dock is.' (Spoiler: We always know where the loading dock is).
To understand this bug, you have to understand how API Platform handles requests. When you hit a standard endpoint, say query { book(id: "/books/1") { ... } }, the system knows exactly what you are doing. It maps the request to a specific Operation on the Book resource. That Operation carries metadata, including security rules like is_granted('ROLE_USER').
However, the Relay node query is polymorphic. It starts at the root. The request is just query { node(id: "...") }. At the moment the query starts, the system doesn't necessarily know what specific resource operation is being performed. It just knows it needs to resolve an ID.
The vulnerability lived in the ResolverFactory. When processing the node field, the system initialized a context for the resolver. In the affected versions, if the operation context wasn't explicitly passed (which is the case for the root node field), the code defaulted to a generic, empty Query object.
Here is the kicker: that generic Query object didn't have your custom security attributes. It was a blank slate. So when the security listener asked, "Are there any rules preventing this user from seeing this?", the generic object replied, "Nope, looks clear to me." The actual resource class (e.g., SecuredDummy) had rules defined, but because the resolver was looking at the generic placeholder operation instead of the specific resource operation, those rules were completely ignored.
Let's look at the smoking gun. The vulnerability was centered in src/GraphQl/Resolver/Factory/ResolverFactory.php. The logic tried to determine which operation was currently being executed to enforce permissions.
In the vulnerable version, the code looked something like this (simplified for clarity):
// Vulnerable Code
$operation = $context['operation'] ?? null;
// If we don't know the operation, just make up a blank one!
$operation ??= new Query();
// ... later ...
if (!$this->security->isGranted($operation->getSecurity(), ...)) {
throw new AccessDeniedException();
}See that ??= new Query()? That is the developer shrugging. Since a fresh new Query() has no security attributes, isGranted checks nothing. It defaults to allowing access because there are no restrictions to enforce.
The fix, applied in commit 60747cc8c2fb855798c923b5537888f8d0969568, forces the system to actually do the homework. Instead of guessing, it now uses a RuntimeOperationMetadataFactory to look up the real operation associated with the IRI.
// Patched Code
try {
// Actually figure out what resource this IRI belongs to
$operation = $this->runtimeOperationMetadataFactory->create($operationName, $resourceClass, true);
} catch (OperationNotFoundException) {
// If we can't find the operation security rules, don't just let them in.
if (null === $operation) {
// Fallback logic that implies strictness or failure
}
}By ensuring the $operation object is hydrated with the metadata from the actual target resource, the security voters are correctly invoked.
Exploiting this is trivially easy if you know the IRI format of the target application. Let's assume we have a resource PrivateDocument that requires ROLE_ADMIN. Normal users should get a 403 Forbidden.
Step 1: The Sanity Check (Failed Access) A standard GraphQL query against the resource works as expected:
query {
privateDocument(id: "/private_documents/1") {
secretContent
}
}Result: "Access Denied." (Because the privateDocument query operation has the security check attached).
Step 2: The Bypass
Now, we switch to the Relay node interface. We ask for the exact same object, but we route the request through the global resolver:
query {
node(id: "/private_documents/1") {
... on PrivateDocument {
id
secretContent
}
}
}Step 3: Execution
Because the node resolver uses the generic Query object (which lacks the ROLE_ADMIN requirement), the API fetches the object from the database and serializes it. The security listener sees an operation with no restrictions and waves it through.
Result:
{
"data": {
"node": {
"id": "/private_documents/1",
"secretContent": "The launch codes are 12345"
}
}
}Here is a visualization of the logic flow:
The impact here is a classic Broken Access Control (CWE-284), but widely distributed. In API Platform, security is often defined centrally on the Resource class (e.g., `#[ApiResource(security:
The maintainers patched this in API Platform Core versions 3.4.17 and 4.0.22. The fix involves a mechanism that properly resolves the underlying resource class from the IRI provided to the node query, ensuring that the correct operation context (and thus the correct security rules) is loaded.
If you are running a vulnerable version, you have two choices:
node query. If your frontend application does not rely on Relay standards, you can disable the node definition in your GraphQL configuration. This removes the attack vector entirely.To developers using API Platform: Do not assume global mechanisms inherit specific constraints. Explicitly test your permissions models using the node query, not just your custom query endpoints.
CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:N/A:N| Product | Affected Versions | Fixed Version |
|---|---|---|
API Platform Core api-platform | >= 4.0.0, < 4.0.22 | 4.0.22 |
API Platform Core api-platform | < 3.4.17 | 3.4.17 |
| Attribute | Detail |
|---|---|
| Vulnerability Type | Authentication Bypass / Broken Access Control |
| CWE ID | CWE-284 |
| CVSS Score | 7.5 (High) |
| Attack Vector | Network |
| Affected Component | GraphQL ResolverFactory (Relay Node) |
| Exploit Status | PoC Available |
Improper Access Control
Grav CMS prior to version 1.7.53 and 2.0.0-rc.8 is vulnerable to an unauthenticated remote denial of service (DoS) vulnerability. By supplying crafted query parameters with extremely large dimensions to image assets, remote unauthenticated attackers can force the server to allocate massive amounts of system memory, leading to kernel Out-Of-Memory (OOM) termination of web worker processes.
CVE-2026-53657 is a local privilege escalation vulnerability in Lima (lima-vm/lima) affecting versions prior to 2.1.3 when configured with the QEMU driver. The guest agent daemon, running as root, creates its communication socket `/run/lima-guestagent.sock` with world-writable permissions (0777). This allows unprivileged local users to command the agent to establish arbitrary tunnels, including to privileged local UNIX sockets (like D-Bus). Because the target daemon authenticates the incoming connection using the credentials of the root-owned guest agent (via SO_PEERCRED), unprivileged users can perform root operations, resulting in complete guest VM compromise.
An unauthenticated Denial of Service vulnerability exists in the s2n-quic library's CryptoStream reassembler due to a lack of buffer limits on out-of-order cryptographic frames. An attacker can transmit a crafted CRYPTO frame with an extremely high offset and nominal payload, forcing the receiver to execute unbounded memory allocations and causing service crashes.
A JNDI Injection and Deserialization Gadget vulnerability exists in mchange-commons-java prior to version 0.6.0. The com.mchange.v2.naming.JavaBeanObjectFactory component permits arbitrary class instantiation and setter invocation, allowing attackers to perform Server-Side Request Forgery (SSRF) and remote class loading.
SurrealDB versions supporting element-level SELECT permissions on arrays are vulnerable to a logical authorization bypass. Due to an index-shifting error during array filtration, restricted elements can skip permission checks and leak to unauthorized record users.
CVE-2026-12243 is a path traversal vulnerability in the Natural Language Toolkit (NLTK) version 3.9.4. The flaw exists because the input validation routine fails to account for percent-encoded directory traversal sequences like '..%2f' before passing them to urllib.request.url2pathname(), which decodes them into active traversal sequences.