MCP server configuration containing stdio/command execution parameters is processed without verifying administrative privileges.
Rule Explorer
Search the public rule index by CVE, GHSA, CWE, language, framework, author, or rule slug. Filter by language, framework, severity, confidence, license, and validation status.
- Public rules
- 4797
- Downloads
- 7.4M
- Verified
- 4797
- Authors
- 2
Usergroup modification without server-side sanitization allows privilege escalation.
Detected direct passing of parsed request body to entity edit logic without filtering sensitive backend configuration properties or checking appropriate permissions.
JWT token claims use a shared audience variable or kernel audience string instead of a scoped service audience, allowing low-privilege service tokens to authenticate against core kernel APIs.
A filterable capability is defined using `apply_filters`, but authorization is checked against hardcoded 'manage_options' rather than the filtered capability variable. This prevents custom capability filters from granting access to non-administrative roles.
A sensitive field like 'role' or 'is_admin' is accepted in validation rules without an explicitly defined allowlist (e.g., `in:`, `Rule::in`). This can lead to vertical privilege escalation via mass assignment if an attacker provides arbitrary administrative values and the backend does not enforce independent authorization for the requested role value.
Instance profile and configuration overrides during migration must enforce project restrictions via project.AllowInstanceUpdate to prevent privilege escalation.
Validated user input is passed directly to staff update service without checking for restricted role assignments like GLOBAL_ADMIN, allowing tenant administrators to escalate privileges.
The GitHub Action workflow runs the `anthropics/claude-code-action` with powerful, unconstrained MCP tools like `mcp__github__submit_pending_pull_request_review`. If the workflow processes pull requests or issues from untrusted authors, this exposes the system to prompt injection. An attacker could hijack the AI reviewer to abuse these tools or exfiltrate th
NewAuthenticator is constructed without a cluster ID resolver parameter. Omitting the cluster ID getter prevents validation of token cluster scopes against the request's cluster ID, potentially leading to authorization bypass across clusters.
prctl(PR_SET_DUMPABLE, 1, ...) is called unconditionally without a boolean guard. In privilege-separated (setuid) architectures where the calling process still holds a privileged IPC channel to a root parent, an attacker sharing the same UID can ptrace-attach to this now-dumpable process and inject arbitrary privileged operations through that channel. Add a
Invite token creation does not verify caller roles or permissions against target invite roles, allowing privilege escalation.
Relying on a user-modifiable configuration dictionary to determine system privileges permits privilege escalation.
A virtual disk backend path is overridden with a locally provided path from `vBD_other_config`. This configuration dictionary is insufficiently restricted and can be modified by lower-privileged administrators (e.g., `vm-admin`). An attacker can exploit this to read or modify arbitrary files in dom0, leading to arbitrary host access or privilege escalation.
Kyverno apiCall constructor invoked without a `policyNamespace` argument. Without this argument, `apiCall.Fetch` substitutes user-controlled context variables into `URLPath` and forwards the resolved path to `client.RawAbsPath` under the Kyverno admission controller ServiceAccount identity, with no enforcement that the request is scoped to the policy's own n
The RoleInterceptor changeActions list is missing 'importOutboundStockMovement', allowing unauthorized low-privilege users to import outbound stock movements.
Impersonation permission is checked against the caller's current evaluation scope rather than against the target context's ACL manager. When changing target contexts or organizations, validate permissions against the target ACL manager to prevent cross-organization privilege escalation.
Impersonation permissions are checked against the caller's global scope instead of evaluating access control using the target organization's ACL manager. This can allow cross-tenant authorization bypass and unauthorized privilege escalation across organization boundaries.
Passing bridge job context into pipeline creation commands may bypass protected branch push permission checks, allowing unauthorized pipeline execution.
Unconfigured LDAP search base falls back to root naming context without logging or restriction. Searching the entire directory root DN for privileged objects like sudo roles allows any user with write access to any LDAP subtree to inject rules and elevate privileges.
FreeIPA Directory Server instance setup is missing the Kerberos principal matching rule configuration.
Policy arguments inherited from another request may retain `DenyOnly = true`. When verifying restricted session policies, this causes only explicit denials to be checked, bypassing explicit allow requirements and leading to privilege escalation. Ensure `DenyOnly = false` is enforced when checking a session policy.
Using `mergo.Merge` with `mergo.WithOverride` to merge a user-supplied `Container` (typed as the upstream `k8s.io/api/core/v1.Container` or reachable through a CR `Spec.Template.Container`) into an operator-built container lets a low-privileged tenant override arbitrary container fields — image, command, args, securityContext (privileged/runAsUser/capabiliti
ClusterRoleBinding (or RoleBinding) grants permissions to the built-in group "system:authenticated" (or "system:unauthenticated"). This binds the referenced role to every authenticated principal in the cluster, effectively making the role's permissions universal. This pattern caused CVE-2025-10725 (opendatahub-operator) by binding a Job/Workload-create role