The finding method is almost as interesting as the bug itself. XBOW is an AI-based offensive security tool, and UAF bugs at library integration points are exactly the kind of thing that slips past human code review — reviewers focus on protocol logic, not on what happens to object lifetimes when a TLS session tears down mid-flight in an error path.

There's a pattern here worth noting: the riskiest attack surfaces in complex C software often aren't in the core logic but at integration boundaries — where one component (Exim) makes assumptions about object lifecycles managed by another (GnuTLS). Those boundaries require simultaneous deep familiarity with both codebases, which is cognitively expensive for humans but maps well to automated analysis.

This is also why "use a well-audited TLS library" doesn't fully transfer safety — you inherit the library's correctness guarantees only for the paths the library authors tested, not for how you call it under load or error conditions.