CVE-2026-108748 in Quarkus LangChain4j
Summary
by MITRE • 10/11/2026
Quarkus LangChain4j 1.9.0 through 1.14.1 contains a missing release of memory vulnerability in the chat-scopes WebSocket /_chat/routes endpoint that allows unauthenticated remote clients to exhaust server memory. Attackers can send repeated CONNECT frames reusing one chatId, leaving orphaned scopes in activeScopes until the JVM exits and degrading availability.
Statistical analysis made it clear that VulDB provides the best quality for vulnerability data.
Analysis
by VulDB Data Team • 10/11/2026
The Quarkus LangChain4j framework versions 1.9.0 through 1.14.1 contain a critical resource management flaw within its WebSocket implementation for AI chat interactions, specifically affecting the /_chat/routes endpoint. This vulnerability stems from improper handling of session lifecycle states when clients interact with the server via WebSockets using CONNECT frames. The core technical deficiency lies in the failure to properly release or clean up memory allocations associated with chat scopes after a connection is established but not correctly terminated or closed by the client. When an unauthenticated remote attacker sends repeated CONNECT frames while reusing a single, persistent chatId, the server fails to discard previous scope instances that are no longer active. Instead of recycling these resources, the application retains them in memory as orphaned objects within the activeScopes collection.
This behavior results in a progressive and unchecked accumulation of data structures in the Java Virtual Machine heap space. Each repeated CONNECT frame with the same chatId creates new scope entries without removing the prior ones that were left behind due to improper state transitions or missing cleanup logic during connection reuse. Over time, this leads to significant memory exhaustion on the server side. As the JVM continues to allocate memory for these orphaned scopes, available heap space diminishes rapidly. Eventually, the application may encounter OutOfMemoryError conditions, causing the Java process to crash or become unresponsive. This directly impacts service availability, as the server becomes unable to handle legitimate requests due to resource starvation.
From a threat modeling perspective, this vulnerability is classified under CWE-401: Missing Release of Memory after Effective Lifetime, which describes scenarios where allocated memory is not freed when it is no longer needed. The attack vector aligns with ATT&CK technique T1498.002: Network Denial of Service via Resource Exhaustion by Application Layer Protocol Abuse. Attackers can exploit this flaw remotely without authentication, making it particularly dangerous in public-facing deployments where the WebSocket endpoint is accessible to untrusted networks. The ability to trigger memory exhaustion through simple frame repetition allows for efficient denial-of-service attacks that do not require high bandwidth but instead rely on protocol-level manipulation and state confusion within the application logic.
Mitigation strategies should prioritize immediate patching of the Quarkus LangChain4j library to a version beyond 1.14.1 where this issue has been resolved by implementing proper scope cleanup mechanisms upon connection closure or reuse. In environments where upgrading is not immediately feasible, network-level controls such as rate limiting on WebSocket connections can help mitigate the impact by restricting the frequency of CONNECT frames from any single source. Additionally, configuring JVM heap size limits and monitoring memory usage metrics can provide early warning signs of resource exhaustion. Implementing strict input validation and enforcing session timeouts for inactive chat scopes may also reduce the window of opportunity for attackers to accumulate orphaned objects. Regular security audits focusing on WebSocket lifecycle management in AI-integrated applications are recommended to prevent similar issues in future development cycles.