CVE-2026-63086
HighCVSS 8.6Exploitation Probability (EPSS)
Low risk24th percentile - higher than 24% of all known CVEs
Summary
In text-generation-inference up to version 3.3.7, a server-side request forgery (SSRF) vulnerability exists in the OpenAI-compatible multimodal chat completions endpoint. An unauthenticated network attacker can force the server to issue arbitrary HTTP GET requests by supplying a crafted image_url value in chat message content.
Risk Assessment
An attacker can scan internal ports, access internal services, and steal credentials from cloud instance metadata endpoints, leading to confidentiality and integrity breaches.
Recommendation
Immediately update text-generation-inference to a version newer than 3.3.7 where this vulnerability is fixed. Until then, restrict access to the multimodal chat completions endpoint and implement validation of image URLs.
Related vulnerabilities
- CVE-2026-78657Critical
The SigmaForms Pro – AI Generated Forms plugin for WordPress up to version 1.4.11 is vulnerable to arbitrary file deletion due to insufficient file path validation in the delete_submission_files function. Unauthenticated attackers can delete arbitrary files on the server, which can easily lead to remote code execution when the right file is deleted (such as wp-config.php). The malicious path traversal URL is submitted via form upload field and stored in the database, with deletion triggered when an administrator deletes the submission record from the admin panel.
- CVE-2026-9055Critical
The Booking for Appointments and Events Calendar – Amelia (Premium) plugin for WordPress versions 8.0 - 9.6.2 is vulnerable to privilege escalation due to insufficient validation of the 'type' parameter in the customer update endpoint. This allows customers to set their role to 'manager' and trigger creation of a WordPress user with the wpamelia-manager role when 'externalId' is set to 0. Unauthenticated attackers can escalate to administrator by first elevating to manager, then creating a provider entity linked to an administrator user ID and overwriting that administrator's password.
- CVE-2026-84699Critical
Team Password Manager before version 14.184.308 fails to enforce authentication requirements in the local account password reset flow. Unauthenticated attackers can reset local account passwords and authenticate as those users to gain unauthorized access.
- CVE-2026-84354Critical
Incorrect authorization in FileSystem in Google Chrome prior to 152.0.7977.75 allowed a remote attacker leveraging social engineering to execute arbitrary code outside the sandbox via a crafted HTML page. (Severity: High)
- CVE-2026-84353Critical
Use after free in Shared Tab Groups in Google Chrome on Android prior to 152.0.7977.75 allowed a remote attacker leveraging social engineering to execute arbitrary code outside the sandbox via a crafted HTML page. (Severity: Critical)
- CVE-2026-84352Critical
Use after free in WebGL in Google Chrome on Android prior to 152.0.7977.75 allowed a remote attacker to execute arbitrary code outside the sandbox via a crafted HTML page. (Severity: Critical)
- CVE-2026-84333Critical
Use-after-free in Dawn in Google Chrome on Android prior to 152.0.7977.75. A remote attacker could execute arbitrary code outside the sandbox via a crafted HTML page.
- CVE-2026-84324Critical
Use after free in Proxy in Google Chrome prior to 152.0.7977.75 allowed a remote attacker to execute arbitrary code outside the sandbox via crafted network traffic. (Chromium security severity: High)
- CVE-2026-84480Critical
WWBN AVideo fails to validate password recovery token expiration in userRecoverPassSave.json.php, allowing attackers to use expired tokens to reset account passwords indefinitely. Attackers who obtain a recovery token can use it at any time to change the target account's password and gain full account access.
- CVE-2026-84479Critical
WWBN AVideo (current e01e41ecc and earlier) makes three login-time security controls depend solely on the client-supplied User-Agent header. The isAVideoEncoder()/isAVideoMobileApp() checks match HTTP_USER_AGENT against a hardcoded literal ("AVideoEncoder"/"AVideoMobileApp") with no IP check or shared secret. An attacker who submits valid credentials and sets User-Agent: AVideoEncoder bypasses two-factor authentication, skips brute-force captcha escalation, and avoids being recorded in the login/device audit history. No patch is available at the time of publication.
Original NVD description (English source)
text-generation-inference through 3.3.7 contains a server-side request forgery (SSRF) vulnerability in the OpenAI-compatible multimodal chat completions endpoint that allows unauthenticated network attackers to coerce the server into issuing arbitrary HTTP GET requests by supplying a crafted image_url value in chat message content. The fetch_image function in router/src/validation.rs performs no validation of private, loopback, link-local, or cloud metadata target addresses, and the reqwest HTTP client follows redirects by default, enabling attackers to bypass scheme checks via redirect chains to reach internal services and cloud instance-metadata endpoints for internal port scanning and credential theft.

