Model Agnosticism and Benchmarking

Software has forever had flaws and humans have forever been finding and fixing them. With LLMs generating code, appsec has also been trying to determine how well LLMs can find flaws. Nico Waisman talks about XBOW's LLM-based pentesting, how it climbed a bug bounty leaderboard, how it uses feedback loops for better pentests, and how they handle (and even welcome!) hallucinations.

In the news, using LLMs to find flaws, directory traversal in an MCP, another resource for learning cloud and AI security, spreadsheets and appsec, and more!

Show Notes: https://securityweekly.com/asw-351