Threat Actors Manipulating LLMs for Automated Vulnerability Exploitation

This report summarizes research showing that LLMs (e.g., GPT-4o, Claude) can be manipulated via an RSA (Role, Scenario, Action) pretexting technique to bypass safety guardrails and automatically generate working exploits against vulnerable Odoo ERP instances, effectively enabling novices to produce attack scripts and increasing the risk to organizations that deploy such software.