# AI Puffer: the plugin's own comment reply becomes stored XSS

> CVE-2026-130121 turns the AI Puffer WordPress plugin's automatic comment reply into a stored cross-site scripting sink.

- **Severity:** Medium: Real and reproducible, but needs unusual setup, user help or a narrow audience. Plan a fix.
- **Category:** Vulnerability
- **Published by:** Patchstack
- **Disclosed:** 2026-11-17
- **CVE:** CVE-2026-130121
- **CVSS:** 7.2
- **Affects:** AI Puffer, GPT3 AI Content Writer, WordPress
- **Research by:** crow
- **Primary source:** https://patchstack.com/database/wordpress/plugin/gpt3-ai-content-generator/vulnerability/wordpress-ai-puffer-mcp-server-for-claude-chatgpt-with-ai-chatbot-plugin-2-5-28-unauthenticated-stored-cross-site-scripting-vulnerability
- **Page:** https://theinjection.dev/items/ai-puffer-comment-reply-stored-xss/

CVE-2026-130121 lets an unauthenticated visitor post a WordPress comment that makes the AI Puffer plugin's auto-reply store HTML unfiltered, where it runs for every reader. Version 2.5.29 fixes it.

## What happened

Patchstack published CVE-2026-130121 on 17 November 2026, a stored cross-site scripting flaw in the WordPress plugin AI Puffer, which is also listed as GPT3 AI Content Writer and uses the slug gpt3-ai-content-generator. Patchstack credits the researcher crow. Every version up to and including 2.5.28 is affected, and 2.5.29 carries the fix.

## How it works

The plugin's insert_comment_reply_logic() function saves the model's generated reply through wp_insert_comment(), which skips the kses filtering WordPress normally applies to comment content. An attacker leaves a comment written so the model repeats an HTML payload in its reply, and that reply is stored unfiltered and executes when any reader loads the page. No account is needed to leave the comment.

## Who is affected

WordPress sites on AI Puffer 2.5.28 or earlier with a Comment Auto-Reply task configured and its Action on reply set to Approve are exposed. The default setting, Hold for moderation, keeps the generated reply unpublished until a human approves it. The attacker's own comment must be approved as well.

## What to do

Update AI Puffer to 2.5.29. Until the update lands, set the Comment Auto-Reply task's Action on reply back to Hold for moderation, and check replies the plugin has already published on pages that accept comments.

## Quick facts

- **CVE:** CVE-2026-130121
- **Affected:** AI Puffer 2.5.28 and earlier (slug gpt3-ai-content-generator)
- **Fixed in:** 2.5.29
- **CVSS:** 7.2, GitHub Advisory Database
- **Precondition:** Comment Auto-Reply set to Approve, not the default Hold for moderation
- **Attack type:** Stored XSS through an AI-generated comment reply
- **Found by:** crow, via Patchstack

## Questions

### Which AI Puffer versions are vulnerable to CVE-2026-130121?

Every AI Puffer release up to and including 2.5.28 is vulnerable to CVE-2026-130121, and version 2.5.29 contains the fix. The plugin ships under the WordPress slug gpt3-ai-content-generator and is also listed as GPT3 AI Content Writer, so a site may show either name in its plugin list when you go looking for it.

### Is my WordPress site exposed if comment auto-reply is off?

Sites with no Comment Auto-Reply task are not exposed to CVE-2026-130121, because the flaw needs the plugin to publish a model-generated reply. Exposure also requires that task's Action on reply to be set to Approve. The default, Hold for moderation, keeps the reply unpublished until a human approves it, which blocks the automatic path.

### How does an attacker trigger the AI Puffer XSS?

An attacker posts a comment crafted so the model repeats an HTML payload in its automatic reply. The plugin stores that reply through wp_insert_comment(), which bypasses the kses filtering WordPress applies to ordinary comment content, so the markup survives and runs for every visitor who loads the page. The attacker's comment must be approved first.

## Sources

1. [Patchstack Advisory](https://patchstack.com/database/wordpress/plugin/gpt3-ai-content-generator/vulnerability/wordpress-ai-puffer-mcp-server-for-claude-chatgpt-with-ai-chatbot-plugin-2-5-28-unauthenticated-stored-cross-site-scripting-vulnerability)
2. [GitHub Advisory](https://github.com/advisories/GHSA-x362-9jrm-3rmc)
