The recent popularity of MCPs is surpassed only by the recent examples deficiencies of their secure design. The most obvious challenge is how MCPs, and many more general LLM use cases, have erased two decades of security principles behind separating code and data. We take a look at how developers are using LLMs to generate code and continue our search for where LLMs are providing value to appsec. We also consider what indicators we'd look for as signs of success. For example, are LLMs driving useful commits to overburdened open source developers? Are LLMs climbing the ranks of bug bounty platforms?

In the news, more examples of prompt injection techniques against LLM features in GitLab and GitHub, the value (and tradeoffs) in rewriting code, secure design lessons from a history of iOS exploitation, checking for all the ways to root, and NIST's approach to (maybe) measuring likely exploited vulns.

Show Notes: https://securityweekly.com/asw-333