CVE-2026-130121

AI Puffer comment replies could publish attacker-influenced script

senols · AI Puffer WordPress plugin

First documented disclosure · Review checkpoint

Summary

AI-generated comment replies could preserve malicious HTML supplied through a visitor’s comment.

Severity and exploitation evidence

High · 7.2/10 · CVSS 3.1. Severity source.

Exploitation not established: Malicious exploitation remains unverified in the reviewed sources.

Exploitation evidence source

Technical severity is not a measure of observed attacks or of risk to your deployment. Attempts, demonstrations, and successful compromise retain the qualifications in the source.

How the weakness works

The reply insertion path bypassed the comment sanitization normally applied by WordPress.

Conditions required

Comment Auto-Reply is configured to approve replies, the attacker’s comment is approved, and the model reproduces the payload. Hold for moderation is the default.

Potential impact

Stored script can execute when another user views the affected page.

Affected versions and configurations

Through 2.5.28.

Fix and mitigation status

2.5.29 removes automatic comment replies and disables saved tasks and queued items.

Practical next steps

Upgrade to 2.5.29 or later and inspect previously posted automated replies. Disable automatic approval until corrected.

Operating lesson

Sanitize model-generated HTML at the publication boundary.

Evidence and limits

Wordfence’s CNA timeline distinguishes vendor notification on November 16 from disclosure on November 17 at 15:12:05 UTC, before the prior review checkpoint. The CVE record was published November 18. The Wordfence page returned an empty response. The undated tagged maintainer changelog confirms removal of automatic replies in 2.5.29 and preservation of previously posted comments: https://plugins.svn.wordpress.org/gpt3-ai-content-generator/tags/2.5.29/readme.txt. No publication date is assigned to that changelog.

Sources