How css‑based clickfix turns ai summarizers into attack vectors

TL;DR: Ai summarizers can be manipulated by hidden css prompts to
output malicious instructions; sanitize inputs, render to visible
plaintext and limit summarization scope.

WHAT’S NEW / WHY IT MATTERS

CloudSEK researchers describe an adaptation of the ClickFix technique
that uses CSS-based obfuscation and ‘‘prompt overdose’’ to make hidden
instructions invisible to humans yet readable by AI summarizers. This
matters because many organizations rely on automated summaries
embedded in email clients, browser extensions and productivity
platforms; those summaries can be treated as authoritative and acted
upon, creating an avenue for social engineering that leads to
ransomware or other operational compromises.

TECHNICAL BREAKDOWN

The approach embeds payloads inside HTML using methods such as
zero‑width characters, white‑on‑white text, tiny font rendering and
off‑screen positioning. Payloads are repeated extensively in hidden
sections so that they occupy a large portion of the model’s context
window — a technique termed “prompt overdose.”…