Danny Avila
Monthly
Access-token theft in LibreChat before 0.8.5 lets a malicious MCP (Model Context Protocol) server hijack OAuth tokens minted for a legitimate server, because the client never checks that the RFC 9728 resource parameter matches the configured MCP server URL. Any LibreChat operator who connects their instance to an attacker-controlled MCP server can have the OAuth access tokens intended for a trusted server silently exfiltrated. SSVC rates technical impact as total and a proof-of-concept is referenced, but there is no public exploit weaponization and EPSS exploitation probability is very low (0.11%); it is fixed in 0.8.5.
Authentication-bypass weakness in LibreChat before 0.8.4-rc1 lets an attacker who already holds a valid session silently regenerate all of a victim's two-factor backup codes via POST /api/auth/2fa/backup/regenerate, which never asks for a current TOTP token or an existing backup code. With fresh codes in hand, the attacker can bypass 2FA at login or turn 2FA off entirely, neutralizing the account's second factor. EPSS is low (0.15%, 5th percentile) and no active exploitation is recorded, but proof-of-concept exploitation is acknowledged in the SSVC framework.
Two-factor authentication takeover in LibreChat before 0.8.4-rc1 lets any authenticated user — or an attacker wielding a stolen session token — call the GET /api/auth/2fa/enable endpoint to silently overwrite an account's existing TOTP secret, regenerate backup codes, and flip twoFactorEnabled to false, all without TOTP or backup-code verification. The flaw destroys the victim's working second factor and can lock the legitimate owner out of their own 2FA. Publicly available exploit code exists (SSVC: PoC), but EPSS is low (0.18%, 8th percentile) and it is not listed in CISA KEV.
Cross-user API key overwrite in LibreChat versions up to and including 0.7.6 allows any authenticated user to replace another user's stored provider API keys (OpenAI, Anthropic, Azure) by injecting a userId parameter into PUT /api/keys requests. The flaw stems from an Insecure Direct Object Reference where the JavaScript spread operator overrides the server-set authenticated user ID, enabling conversation hijacking or denial of service. No public exploit identified at time of analysis, and the issue is fixed in 0.8.3-rc1.
Server-side request forgery in LibreChat 0.8.2-rc2 through 0.8.2 allows authenticated users to access internal network resources via incomplete DNS validation bypass. Despite a prior SSRF patch, the current hostname validation fails to check if DNS resolution points to private IP addresses, enabling attackers to reach internal RAG APIs and cloud metadata endpoints. CVSS 7.7 with network-based attack vector and low complexity. EPSS data not available; no confirmed active exploitation (not listed in CISA KEV). Patch released in version 0.8.3-rc1.
Server-Side Request Forgery (SSRF) in LibreChat versions prior to 0.8.3 allows authenticated users to bypass IP validation and force the application server to make HTTP requests to internal network resources. The vulnerability stems from improper validation of IPv4-mapped IPv6 addresses in hex-normalized form, enabling access to cloud metadata services (AWS 169.254.169.254), loopback addresses, and RFC1918 private networks. With EPSS data unavailable and no CISA KEV listing, no public exploit identified at time of analysis, though the specific bypass technique (hex-normalized IPv4-mapped IPv6) is well-documented in SSRF research.
LibreChat versions 0.8.2 through 0.8.2-rc3 contain an authentication bypass vulnerability in the Model Context Protocol (MCP) OAuth callback endpoint that allows attackers to steal OAuth tokens by tricking victims into completing an OAuth flow, resulting in account takeover of the victim's MCP-linked services like Atlassian and Outlook. No active exploitation is known (not in KEV), no POC is publicly available, and EPSS data is not yet available for this newly disclosed vulnerability.
Unrestricted file upload to the POST /api/convos/import endpoint in LibreChat prior to 0.8.4-rc1 allows any authenticated user to exhaust server disk space and memory, causing a denial of service. The root cause is an incomplete remediation of CVE-2024-11171: while file size limits were added to the shared multer factory, the import endpoint's independent multer instance was never updated, and the application-level fallback guard (CONVERSATION_IMPORT_MAX_FILE_SIZE_BYTES) is commented out in .env.example by default. No public exploit code or CISA KEV listing has been identified at time of analysis, but the low exploitation complexity for any authenticated user makes this a meaningful operational risk for public-facing or open-registration deployments.
Cross-site scripting in LibreChat's markdown artifact preview pipeline allows an authenticated attacker to inject arbitrary JavaScript into a victim's browser session via crafted image alt text. The marked library v15.0.12 fails to HTML-escape double-quote characters in image alt text when LibreChat's custom image renderer falls through to the built-in renderer, enabling attribute breakout via payloads like `" onload="payload`. The resulting unsanitized HTML is written directly to `innerHTML` inside the Sandpack preview iframe, executing attacker-controlled code in the victim's browser. No public exploit code or active exploitation (CISA KEV) has been identified at time of analysis; fixed in version 0.8.4-rc1.
LibreChat's POST /api/files/images endpoint lacks the authorization check that was correctly applied to the POST /api/files route in a prior patch, allowing any authenticated user to upload files into arbitrary agents' tool_resources (including context and execute_code environments) without ownership or EDIT permission verification. All LibreChat deployments prior to 0.8.4-rc1 are affected across any multi-user configuration. No active exploitation has been confirmed (not in CISA KEV), but the bypass is trivially exercisable by any valid session holder who knows a target agent ID.
Broken object-level authorization in LibreChat's message deletion endpoint allows any authenticated user to permanently delete messages belonging to other users. Affected are all LibreChat deployments prior to 0.8.4-rc1. By supplying their own valid conversationId for middleware validation while targeting a victim's messageId in the underlying MongoDB query, an attacker bypasses the per-user authorization check and irreversibly removes arbitrary messages. No public exploit or CISA KEV listing exists at time of analysis, but the flaw is straightforward to exploit once a valid messageId is obtained.
Server-side request forgery in LibreChat before 0.8.4-rc1 lets any authenticated user point a custom OpenAI-compatible endpoint baseURL at internal network addresses, because the URL is used to build HTTP requests with no SSRF validation whatsoever - no private-IP blocking, no scheme restriction, and no DNS pinning. By abusing the legitimate custom-endpoint feature, a logged-in user can coerce the server into reaching internal-only services and returning their responses, with a CVSS 3.1 score of 7.7 driven by a changed scope and high confidentiality impact. There is no public exploit identified at time of analysis and the issue is not in CISA KEV, but the trivial low-complexity nature makes it readily exploitable once an account exists.
Rate limiter bypass in LibreChat prior to 0.8.4-rc1 allows any authenticated user to exhaust server resources via the POST /api/convos/duplicate endpoint, which performs identical expensive database operations as /api/convos/fork but was inadvertently omitted from the CVE-2025-7105 remediation scope. Operators who applied the prior fix may believe resource exhaustion was fully mitigated - they remain exposed through this overlooked parallel endpoint. No public exploit code has been identified at time of analysis and the vulnerability is not listed in CISA KEV, but the low exploitation complexity and bypass nature make this operationally significant for multi-tenant or publicly accessible LibreChat deployments.
Arbitrary file write in LibreChat prior to 0.8.4 allows authenticated users to overwrite arbitrary server files via path traversal in code artifact filenames. The vulnerability affects LibreChat deployments using the default local file storage strategy, where the execute_code sandbox returns a user-controllable filename that is concatenated directly into the file write path without sanitization. An authenticated attacker can craft malicious artifact names containing traversal sequences (e.g., ../../../../../app/client/dist/poc.txt) to write files outside the intended directory, potentially compromising application integrity or enabling remote code execution through client-side file injection.
LibreChat versions 0.8.2-rc1 through 0.8.3-rc1 leak OAuth access tokens when authenticated users interact with malicious MCP servers, which can inject credential placeholders into HTTP headers that are automatically substituted with sensitive tokens. An attacker can create a rogue MCP server containing headers like {{LIBRECHAT_OPENID_ACCESS_TOKEN}} to harvest victim credentials during tool execution; the vulnerability is fixed in version 0.8.3-rc2. No public exploit code or CISA KEV listing is documented, but the attack requires user interaction (UI:R) and authenticated access (PR:L), limiting real-world impact.
Unauthenticated stream hijacking in LibreChat versions 0.8.2-rc2 through 0.8.2-rc3 allows authenticated users to read other users' real-time chat conversations via the SSE streaming endpoint `/api/agents/chat/stream/:streamId` without ownership verification. An attacker with valid credentials can enumerate or guess stream IDs to intercept sensitive messages, AI-generated responses, and tool invocation data from arbitrary users. The vulnerability was patched in version 0.8.2.
A critical authentication bypass vulnerability exists in LibreChat version 0.8.1-rc2 where the same JWT secret is reused for both user session management and the RAG (Retrieval-Augmented Generation) API authentication. This design flaw allows authenticated users to compromise service-level authentication of the RAG API by leveraging their session tokens to access or manipulate the RAG service beyond intended privileges. No active exploitation (KEV) has been reported, but a detailed security advisory with technical analysis is publicly available from SBA Research.
LibreChat versions prior to 0.8.3-rc1 contain a Denial of Service vulnerability in the DELETE /api/convos endpoint where authenticated attackers can crash the Node.js server process by sending malformed requests lacking the required req.body.arg parameter. The vulnerability exploits improper destructuring without validation, causing an unhandled TypeError that bypasses Express middleware and triggers process.exit(1), resulting in complete service unavailability. No evidence of active exploitation in the wild or public POC has been identified at this time.
Access-token theft in LibreChat before 0.8.5 lets a malicious MCP (Model Context Protocol) server hijack OAuth tokens minted for a legitimate server, because the client never checks that the RFC 9728 resource parameter matches the configured MCP server URL. Any LibreChat operator who connects their instance to an attacker-controlled MCP server can have the OAuth access tokens intended for a trusted server silently exfiltrated. SSVC rates technical impact as total and a proof-of-concept is referenced, but there is no public exploit weaponization and EPSS exploitation probability is very low (0.11%); it is fixed in 0.8.5.
Authentication-bypass weakness in LibreChat before 0.8.4-rc1 lets an attacker who already holds a valid session silently regenerate all of a victim's two-factor backup codes via POST /api/auth/2fa/backup/regenerate, which never asks for a current TOTP token or an existing backup code. With fresh codes in hand, the attacker can bypass 2FA at login or turn 2FA off entirely, neutralizing the account's second factor. EPSS is low (0.15%, 5th percentile) and no active exploitation is recorded, but proof-of-concept exploitation is acknowledged in the SSVC framework.
Two-factor authentication takeover in LibreChat before 0.8.4-rc1 lets any authenticated user — or an attacker wielding a stolen session token — call the GET /api/auth/2fa/enable endpoint to silently overwrite an account's existing TOTP secret, regenerate backup codes, and flip twoFactorEnabled to false, all without TOTP or backup-code verification. The flaw destroys the victim's working second factor and can lock the legitimate owner out of their own 2FA. Publicly available exploit code exists (SSVC: PoC), but EPSS is low (0.18%, 8th percentile) and it is not listed in CISA KEV.
Cross-user API key overwrite in LibreChat versions up to and including 0.7.6 allows any authenticated user to replace another user's stored provider API keys (OpenAI, Anthropic, Azure) by injecting a userId parameter into PUT /api/keys requests. The flaw stems from an Insecure Direct Object Reference where the JavaScript spread operator overrides the server-set authenticated user ID, enabling conversation hijacking or denial of service. No public exploit identified at time of analysis, and the issue is fixed in 0.8.3-rc1.
Server-side request forgery in LibreChat 0.8.2-rc2 through 0.8.2 allows authenticated users to access internal network resources via incomplete DNS validation bypass. Despite a prior SSRF patch, the current hostname validation fails to check if DNS resolution points to private IP addresses, enabling attackers to reach internal RAG APIs and cloud metadata endpoints. CVSS 7.7 with network-based attack vector and low complexity. EPSS data not available; no confirmed active exploitation (not listed in CISA KEV). Patch released in version 0.8.3-rc1.
Server-Side Request Forgery (SSRF) in LibreChat versions prior to 0.8.3 allows authenticated users to bypass IP validation and force the application server to make HTTP requests to internal network resources. The vulnerability stems from improper validation of IPv4-mapped IPv6 addresses in hex-normalized form, enabling access to cloud metadata services (AWS 169.254.169.254), loopback addresses, and RFC1918 private networks. With EPSS data unavailable and no CISA KEV listing, no public exploit identified at time of analysis, though the specific bypass technique (hex-normalized IPv4-mapped IPv6) is well-documented in SSRF research.
LibreChat versions 0.8.2 through 0.8.2-rc3 contain an authentication bypass vulnerability in the Model Context Protocol (MCP) OAuth callback endpoint that allows attackers to steal OAuth tokens by tricking victims into completing an OAuth flow, resulting in account takeover of the victim's MCP-linked services like Atlassian and Outlook. No active exploitation is known (not in KEV), no POC is publicly available, and EPSS data is not yet available for this newly disclosed vulnerability.
Unrestricted file upload to the POST /api/convos/import endpoint in LibreChat prior to 0.8.4-rc1 allows any authenticated user to exhaust server disk space and memory, causing a denial of service. The root cause is an incomplete remediation of CVE-2024-11171: while file size limits were added to the shared multer factory, the import endpoint's independent multer instance was never updated, and the application-level fallback guard (CONVERSATION_IMPORT_MAX_FILE_SIZE_BYTES) is commented out in .env.example by default. No public exploit code or CISA KEV listing has been identified at time of analysis, but the low exploitation complexity for any authenticated user makes this a meaningful operational risk for public-facing or open-registration deployments.
Cross-site scripting in LibreChat's markdown artifact preview pipeline allows an authenticated attacker to inject arbitrary JavaScript into a victim's browser session via crafted image alt text. The marked library v15.0.12 fails to HTML-escape double-quote characters in image alt text when LibreChat's custom image renderer falls through to the built-in renderer, enabling attribute breakout via payloads like `" onload="payload`. The resulting unsanitized HTML is written directly to `innerHTML` inside the Sandpack preview iframe, executing attacker-controlled code in the victim's browser. No public exploit code or active exploitation (CISA KEV) has been identified at time of analysis; fixed in version 0.8.4-rc1.
LibreChat's POST /api/files/images endpoint lacks the authorization check that was correctly applied to the POST /api/files route in a prior patch, allowing any authenticated user to upload files into arbitrary agents' tool_resources (including context and execute_code environments) without ownership or EDIT permission verification. All LibreChat deployments prior to 0.8.4-rc1 are affected across any multi-user configuration. No active exploitation has been confirmed (not in CISA KEV), but the bypass is trivially exercisable by any valid session holder who knows a target agent ID.
Broken object-level authorization in LibreChat's message deletion endpoint allows any authenticated user to permanently delete messages belonging to other users. Affected are all LibreChat deployments prior to 0.8.4-rc1. By supplying their own valid conversationId for middleware validation while targeting a victim's messageId in the underlying MongoDB query, an attacker bypasses the per-user authorization check and irreversibly removes arbitrary messages. No public exploit or CISA KEV listing exists at time of analysis, but the flaw is straightforward to exploit once a valid messageId is obtained.
Server-side request forgery in LibreChat before 0.8.4-rc1 lets any authenticated user point a custom OpenAI-compatible endpoint baseURL at internal network addresses, because the URL is used to build HTTP requests with no SSRF validation whatsoever - no private-IP blocking, no scheme restriction, and no DNS pinning. By abusing the legitimate custom-endpoint feature, a logged-in user can coerce the server into reaching internal-only services and returning their responses, with a CVSS 3.1 score of 7.7 driven by a changed scope and high confidentiality impact. There is no public exploit identified at time of analysis and the issue is not in CISA KEV, but the trivial low-complexity nature makes it readily exploitable once an account exists.
Rate limiter bypass in LibreChat prior to 0.8.4-rc1 allows any authenticated user to exhaust server resources via the POST /api/convos/duplicate endpoint, which performs identical expensive database operations as /api/convos/fork but was inadvertently omitted from the CVE-2025-7105 remediation scope. Operators who applied the prior fix may believe resource exhaustion was fully mitigated - they remain exposed through this overlooked parallel endpoint. No public exploit code has been identified at time of analysis and the vulnerability is not listed in CISA KEV, but the low exploitation complexity and bypass nature make this operationally significant for multi-tenant or publicly accessible LibreChat deployments.
Arbitrary file write in LibreChat prior to 0.8.4 allows authenticated users to overwrite arbitrary server files via path traversal in code artifact filenames. The vulnerability affects LibreChat deployments using the default local file storage strategy, where the execute_code sandbox returns a user-controllable filename that is concatenated directly into the file write path without sanitization. An authenticated attacker can craft malicious artifact names containing traversal sequences (e.g., ../../../../../app/client/dist/poc.txt) to write files outside the intended directory, potentially compromising application integrity or enabling remote code execution through client-side file injection.
LibreChat versions 0.8.2-rc1 through 0.8.3-rc1 leak OAuth access tokens when authenticated users interact with malicious MCP servers, which can inject credential placeholders into HTTP headers that are automatically substituted with sensitive tokens. An attacker can create a rogue MCP server containing headers like {{LIBRECHAT_OPENID_ACCESS_TOKEN}} to harvest victim credentials during tool execution; the vulnerability is fixed in version 0.8.3-rc2. No public exploit code or CISA KEV listing is documented, but the attack requires user interaction (UI:R) and authenticated access (PR:L), limiting real-world impact.
Unauthenticated stream hijacking in LibreChat versions 0.8.2-rc2 through 0.8.2-rc3 allows authenticated users to read other users' real-time chat conversations via the SSE streaming endpoint `/api/agents/chat/stream/:streamId` without ownership verification. An attacker with valid credentials can enumerate or guess stream IDs to intercept sensitive messages, AI-generated responses, and tool invocation data from arbitrary users. The vulnerability was patched in version 0.8.2.
A critical authentication bypass vulnerability exists in LibreChat version 0.8.1-rc2 where the same JWT secret is reused for both user session management and the RAG (Retrieval-Augmented Generation) API authentication. This design flaw allows authenticated users to compromise service-level authentication of the RAG API by leveraging their session tokens to access or manipulate the RAG service beyond intended privileges. No active exploitation (KEV) has been reported, but a detailed security advisory with technical analysis is publicly available from SBA Research.
LibreChat versions prior to 0.8.3-rc1 contain a Denial of Service vulnerability in the DELETE /api/convos endpoint where authenticated attackers can crash the Node.js server process by sending malformed requests lacking the required req.body.arg parameter. The vulnerability exploits improper destructuring without validation, causing an unhandled TypeError that bypasses Express middleware and triggers process.exit(1), resulting in complete service unavailability. No evidence of active exploitation in the wild or public POC has been identified at this time.