By weakness (CWE)

CWE-915: related vulnerabilities

CVEs classified under CWE-915. Understanding the weakness class helps prioritize systemic fixes over one-off patches.

20 published vulnerabilities

• CVE-2026-46475HIGH 8.8

  Flowise, a platform for building customized LLM workflows through a drag-and-drop interface, contains a privilege escalation vulnerability in its assistant management features. Prior to version 3.1.2, an authenticated attacker can exploit mass-assignment flaws in the create and update endpoints to take over assistants belonging to other workspaces, potentially accessing or modifying shared LLM configurations across organizational boundaries.

• CVE-2026-46476HIGH 8.8

  Flowise, a visual tool for building customized large language model workflows, contains a vulnerability in how it handles template creation and updates. Attackers with user access can exploit mass-assignment flaws to take control of templates across different workspaces—essentially hijacking template configurations that should be isolated from one another. The vulnerability affects all versions prior to 3.1.2 and has been resolved in that release.

• CVE-2026-46477HIGH 8.8

  Flowise, a visual interface for building and customizing language model workflows, contains a mass-assignment vulnerability in its dataset management functions. Before version 3.1.2, an authenticated attacker could exploit weak input validation during dataset creation or updates to access and modify datasets across different workspaces—effectively taking over datasets belonging to other users or teams. The vulnerability requires a valid user account to exploit but poses a serious risk to multi-tenant Flowise deployments where data isolation is critical.

• CVE-2026-46478HIGH 8.8

  Flowise, a popular drag-and-drop interface for building customized language model workflows, contains a privilege escalation flaw in its DatasetRow functionality. Before version 3.1.2, an authenticated attacker could manipulate how new dataset rows are created or updated, allowing them to inject or modify rows belonging to other workspaces. This mass-assignment vulnerability effectively grants an attacker control over another organization's data within the same Flowise deployment, provided they have valid login credentials.

• CVE-2026-46479HIGH 8.8

  Flowise, a no-code platform for building customized LLM workflows, contains a privilege escalation vulnerability in its evaluation management system. Attackers with valid user credentials can modify evaluation records in ways that bypass workspace isolation, allowing them to view, edit, or delete evaluations belonging to other teams or organizations. The vulnerability stems from improper input validation in the create and update endpoints—a classic mass-assignment flaw. Version 3.1.2 and later patch this issue.

• CVE-2026-46480HIGH 8.8

  Flowise, a no-code platform for building customized large language model workflows, contains a privilege escalation vulnerability in its evaluator management feature. An authenticated attacker can exploit improper input validation during evaluator creation or updates to gain unauthorized access to evaluators across different workspaces. This allows an attacker to take over evaluators belonging to other users or teams, potentially manipulating AI workflow logic and data without authorization. The vulnerability requires an existing login but no elevated privileges to exploit.

• CVE-2026-44494HIGH 8.7

  Axios, a widely-used HTTP client library for JavaScript applications, contains a vulnerability that allows attackers to intercept and modify all network traffic from an affected application. An attacker who can pollute an application's prototype chain (through a separate vulnerability in a dependency) can inject a malicious proxy configuration into Axios. Once injected, Axios will route all HTTP requests—including those carrying authentication credentials—through the attacker's proxy server, enabling complete man-in-the-middle attacks. This affects Axios versions 1.0.0 through 1.15.x; upgrading to 1.16.0 or later eliminates the vulnerability.

• CVE-2026-45687HIGH 8.5

  Rocket.Chat versions prior to specific patch releases contain a file upload vulnerability that allows authenticated users to manipulate their own uploaded file records in ways the application never intended. When a user uploads a file, the application fails to validate which fields can be modified, allowing an attacker to rewrite metadata about where files are stored and how they're referenced in the system. This could lead to unauthorized access to file storage systems or manipulation of file locations, though the vulnerability requires an existing user account to exploit.

• CVE-2026-54351HIGH 8.2

  Budibase versions before 3.39.9 contain a vulnerability in their webhook trigger system that allows an authenticated attacker to hijack automations. By crafting a malicious webhook POST request with a specially modified body, an attacker can trick the system into executing an automation under a different application context than intended. This grants the attacker unauthorized access to read and modify data in the victim's workspace database. The flaw stems from the webhook endpoint accepting and passing unvalidated user input directly into automation parameters without proper access controls.

• CVE-2026-42863HIGH 8.1

  FlowiseAI versions prior to 3.1.2 contain a mass assignment vulnerability in their chatflow update feature. An authenticated user can modify internal system properties—such as workspace assignment, deployment status, and visibility settings—that should only be controlled by administrators. This allows attackers to reassign workflows to other workspaces, change deployment states without authorization, and alter public/private visibility of chatflows. The vulnerability requires valid login credentials but no additional special access.

• CVE-2026-46517HIGH 7.8

  LMDeploy, a toolkit for compressing and deploying large language models, contains a critical flaw in versions 0.12.3 and earlier. The tool automatically trusts code downloaded from Hugging Face without asking users for permission. This "trust by default" setting allows attackers to inject malicious code into the supply chain—if a model or dependency on Hugging Face is compromised, LMDeploy will execute that code automatically on any system using the toolkit. No patches are currently available.

• CVE-2026-46625HIGH 7.5

  js-cookie is a popular JavaScript library that manages browser cookies. Versions before 3.0.7 contain a prototype pollution vulnerability where an attacker can manipulate cookie security attributes. By crafting a malicious JSON object with a specially-formed __proto__ property, an attacker can inject unauthorized cookie attributes like domain, secure, samesite, expires, and path. This allows them to broaden cookie scope, remove security restrictions, or extend cookie lifetime—potentially enabling session hijacking, cross-site request forgery, or credential theft depending on how the application uses cookies.

• CVE-2026-59721HIGH 7.2

  Hoppscotch, an open-source API development platform, contains a critical configuration vulnerability that allows administrators with legitimate access to inadvertently or maliciously enable arbitrary command execution on the backend server. By crafting a specially formatted SMTP URL in the infrastructure settings, an attacker with admin privileges can inject sendmail transport options that execute system commands with root privileges when the service restarts and attempts to send mail. This is a post-authentication attack requiring admin access, but the impact—root code execution on the backend—is severe.

• CVE-2026-44495HIGH 7.0

  Axios, a widely-used HTTP client library for JavaScript applications, contains prototype-pollution gadgets that can be exploited if another vulnerability in the same process has already poisoned JavaScript's Object.prototype. The issue affects versions 0.19.0 through 0.31.0 in the 0.x branch and version 1.15.1 in the 1.x branch. By itself, Axios does not cause the initial prototype pollution; instead, it becomes a secondary target once an attacker has already compromised Object.prototype through a separate vulnerability. When this occurs, Axios may incorrectly interpret inherited properties as legitimate request configuration, leading to potential code execution or data exposure. The vulnerability is addressed in Axios 0.31.1 and 1.15.2.

• CVE-2026-48943MEDIUM 6.5

  K2, a popular content component for Joomla, has a flaw in how it handles user profile updates. When a registered Joomla user submits a profile edit through the standard Joomla user management system, they can inject a special parameter that tricks K2 into letting them modify hidden database fields—specifically their notes, profile image, and plugin settings—that the normal K2 interface doesn't expose. This allows unauthorized modification of user profile data that should be restricted to administrators.

• CVE-2026-54601MEDIUM 6.3

  FastGPT, an open-source AI knowledge base platform, contains a cross-tenant data isolation flaw affecting versions 4.14.17 through 4.15.0-beta3. An authenticated user in one tenant can exploit a dataset collection creation endpoint to gain unauthorized access to datasets belonging to another tenant. Once the attacker knows or discovers dataset identifiers from another tenant, they can read, modify, or delete that data. The vulnerability stems from inconsistent authorization logic that trusts ownership information from mixed or corrupted dataset objects rather than validating against the calling user's actual tenant context.

• CVE-2026-54515MEDIUM 5.3

  Jackson-databind, a widely-used Java library for converting JSON to Java objects, has a flaw that allows properties marked to be ignored during deserialization to become writable again in certain configurations. When a developer uses @JsonIgnoreProperties to exclude specific fields from being deserialized, and also enables case-insensitive property matching via @JsonFormat, the library incorrectly rebuilds its property filter from the unfiltered map, reinstating the very properties it should block. An attacker who can supply JSON input to an affected application could write to fields that should be protected, potentially modifying application state or bypassing business logic constraints.

• CVE-2026-54516MEDIUM 5.3

  Jackson-databind, a widely-used Java library for converting JSON to and from objects, has a flaw in how it handles property annotations. The bug allows attackers to write data to fields that should be protected by @JsonIgnore annotations when certain conditions are met. Specifically, if a class uses @JsonProperty to rename a getter but @JsonIgnore on the setter, an attacker can bypass the ignore directive by sending JSON with the renamed key, directly modifying the underlying field. This works because Jackson's property inference logic treats the backing field as writable even though the intended API forbids it. The vulnerability affects versions 2.21.0 through 2.21.4 and 3.1.4 (or earlier in the 3.x line prior to 3.1.4).

• CVE-2026-42862MEDIUM 5.0

  Flowise, a popular drag-and-drop interface for building custom AI language model workflows, contains a security flaw that allows authenticated users to move tools between workspaces without proper authorization. When updating a tool, the application fails to validate who should have permission to change ownership fields like workspaceId. An attacker with legitimate access to one workspace can reassign tools to a different workspace, potentially exposing or stealing AI workflows, data pipelines, or proprietary configurations belonging to another team or customer. This breaks the isolation that multi-workspace Flowise deployments rely on to keep organizations separate.

• CVE-2026-42540MEDIUM 4.3

  IRIS is a collaborative web platform used by incident response teams to share and document technical details during security investigations. A vulnerability in versions before 2.4.28 allows authenticated users to modify database records through specially crafted API requests, potentially corrupting or altering incident investigation data. The issue requires valid login credentials to exploit and affects data integrity rather than confidentiality.