The enemy within: AI as the attack surface

Tenable researchers have revealed a set of vulnerabilities, dubbed "HackedGPT," that exploit the expanded attack surface created by large-language models (LLMs) and AI assistants, particularly through techniques like indirect prompt injection and malicious web content. These exploits can enable data exfiltration, malware persistence, and unauthorized access by manipulating AI systems that browse live websites, remember user context, or connect to business applications, highlighting significant security risks. The findings underscore the necessity for rigorous governance, controls, and monitoring of AI operations, treating AI systems as critical user or device entities subject to strict audit protocols. While some