VibeScamming refers to using AI for generating phishing scams with minimal skill required. Guardio Labs developed the VibeScamming Benchmark to evaluate AI resistance against misuse in phishing workflows. The benchmark's first round tested ChatGPT, Claude, and Lovable, highlighting their varying degrees of resilience. The results reveal important insights into the balance between AI capabilities and their potential abuses, stressing the need for robust safety measures.
Table of contents
VibeScamming — From Prompt to Phish: Benchmarking Popular AI Agents’ Resistance to the Dark SideThe Future of Scamming is Already HereBenchmark GuidelinesTest Subjects — From Top Players to NewcomersPart I — Incepting the IdeaPart II — From POC to Production1. The Visual Similarity Test2. Mitigations — Hiding from Detection3. Hosting — Going Live4. C2 — Collecting Credentials5. Messaging — Reaching Your VictimsThe Results Are InSummary57 Impressions