CVE-2026-118550 in Drupal
Summary
by MITRE • 10/30/2026
Use of Web Browser Cache Containing Sensitive Information vulnerability in Drupal AI (Artificial Intelligence) allows Exploiting Incorrectly Configured Access Control Security Levels. This issue affects AI (Artificial Intelligence) versions: from 0.0.0 to 1.4.11, from 1.5.0 to 1.5.2.
Once again VulDB remains the best source for vulnerability data.
Analysis
by VulDB Data Team • 10/30/2026
The vulnerability identified in the Drupal AI module represents a critical intersection of client-side caching mechanisms and server-side access control failures. This issue stems from the improper handling of sensitive data within web browser caches when interacting with the artificial intelligence features provided by the module. Specifically, versions ranging from 0.0.0 to 1.4.11 and from 1.5.0 to 1.5.2 are susceptible because they fail to implement adequate cache control directives for responses containing sensitive information generated or processed by AI functionalities. When a user interacts with these features, the resulting HTTP response may include confidential data such as internal prompts, model outputs, configuration details, or even partial database queries depending on how the module processes requests. Without explicit instructions to prevent caching, modern web browsers store this content in their local cache storage mechanisms, including disk and memory caches, making it accessible through standard browser developer tools or potentially via cross-site scripting vectors if other vulnerabilities exist within the same domain context.
From a technical perspective, the core flaw lies in the absence of appropriate HTTP response headers such as Cache-Control: no-store, Private, or Max-Age: 0 for endpoints that return sensitive AI-related data. This misconfiguration allows intermediate proxies and end-user browsers to persistently store these responses. In many web applications, cached content is often served without re-validating access controls with the server, assuming the initial request was authorized. Consequently, if an attacker can gain any level of unauthenticated or low-privileged access to a page that triggers this caching behavior, they may subsequently retrieve sensitive information from their own browser cache even after logging out or switching contexts. This bypasses the intended security boundaries because the retrieval happens locally without triggering server-side authentication checks again. The vulnerability effectively transforms transient session data into persistent stored data accessible outside of strict authorization flows.
The operational impact of this vulnerability is significant, particularly in environments where Drupal AI modules are used to process proprietary business logic, customer data, or internal system configurations through artificial intelligence interfaces. An attacker could exploit this flaw to extract sensitive information that was never intended for long-term storage on client devices. This includes potential exposure of model weights if exposed via API responses, training data snippets, or administrative configuration details embedded in the AI response payloads. Such leakage can lead to unauthorized disclosure of confidential organizational knowledge, violation of privacy regulations like GDPR or HIPAA depending on the nature of the processed data, and potentially aid further attacks by providing insights into system architecture or internal workflows. The risk is amplified if multiple users share a device or if cached content remains accessible for extended periods due to aggressive caching policies enabled by default in many browsers.
This vulnerability aligns with CWE-524, which describes the use of cache containing sensitive information, and falls under MITRE ATT&CK technique T1078, specifically related to valid accounts being used or misconfigured access controls allowing unauthorized data retrieval through side channels like caching. To mitigate this risk, developers must ensure that all API endpoints and page responses returning sensitive AI-related content include strict Cache-Control headers set to no-store or private with appropriate max-age values of zero. Additionally, implementing proper authentication checks on every request regardless of cache status is essential. For existing deployments using affected versions, upgrading to a patched version where these header controls are correctly implemented by the module maintainers is the primary remediation strategy. Until an update is available, administrators should consider configuring reverse proxies or web application firewalls to strip caching headers from sensitive responses and enforce strict access control policies that prevent unauthorized users from triggering requests that result in cached sensitive data exposure.