Now You See Me, Now You Don’t: Using LLMs to Obfuscate Malicious Java Script
Summary: Unit 42 developed an adversarial machine learning algorithm that uses LLMs to iteratively rewrite malicious JavaScript, flipping a deep learning model's verdict from malicious to benign 88% of the time across a few hundred unique samples. In one example, stacked minification, string splitting and variable renaming dropped the phishing detection score from nearly 1.0 to 0.007577492, and no VirusTotal vendors detected the rewritten script as malicious even after four days.
Key Takeaway: Retraining the classifier on 10,000 LLM-rewritten samples raised real-world detection on 2022-and-later samples by 10%, which means detection gains now depend on defenders generating the same adversarial variants attackers can produce at scale.
Read the original at Unit 42 Palo Alto Networks ↗
- Topics
- Zero-Days & Exploits
- Related
- New Sign Drop Exploit Can Abuse Windows Defender to Bypass Security Protections Cyber Risks · Nov 13, 2026New Sign Drop Zero-Day Exploit Turns Windows CLFS Security Feature Into SYSTEM Access Cyber Risks · Nov 12, 2026November Patch Tuesday: An Overflowing Cornucopia of CVEs Cyber Risks · Nov 10, 2026