I have a short piece in the latest issue of the CAS Actuarial Review on “Vibe Coding and the Next Silent Cyber Risk.” As AI-generated code becomes part of ordinary software development, the challenge for insurers may not simply be whether the code is more or less secure. It is also a question of provenance: after a loss, can we reconstruct what was prompted, generated, reviewed, tested, and ultimately deployed? In a nutshell, no provenance, no confidence.