AI vs AI: Creating Novel Spam and Catching it with Text Generating AI

GPT is a powerful text generation model, but its text generation is unconstrained. This session will discuss how to control GPT to produce sentences which meet specific style or content requirements. The session will review approaches that allow attendees to drive GTP to write SPAM and HAM messages and will demonstrate how to convert GPT into a SPAM filter which detects the generated SPAM.

Younghoo Lee

CatBERT: Context-Aware Tiny BERT for Detecting Targeted Social Engineering Emails

Targeted phishing emails are a major cyber threat on the Internet today and are insufficiently addressed by current defenses. In this paper, we leverage industrial-scale datasets from Sophos cloud email security service, which defends tens of millions of customer mailboxes, to propose a novel Transformer-based architecture for detecting targeted phishing emails. Our model leverages both natural language and email header inputs, is more computationally efficient than competing transformer approaches, and we show that it is less prone to adversarial attacks which deliberately replace keywords with typos or synonyms.

Younghoo Lee
Joshua Saxe
Richard Harang