XBOW's Nico Waisman on who owns AI code risk when models drift
Key takeaways from theCUBE AppDevANGLE interview with XBOW CISO Nico WaismanDealroom has a profile for this one. Try Dealroom → (November 2026; transcript unavailable, based on episode description):
AI code risk. AI-generated code is entering production faster than enterprise security postures can absorb it; AI-assisted development is now standard practice rather than an experiment.
Benchmark decay. LLM-based security products face benchmark decay — model upgrades can silently change application behaviour.
Safety by engineering. Safety in AI-offensive tools requires engineering infrastructure around the model, not prompt-level controls.
Paved roads. A paved-road model structures incentives so the secure path is also the fastest path for development teams and AI agents.
Exploitation over inventory. Exploitation potential, not vulnerability inventory, is the more reliable signal for prioritising remediation.
Build versus buy. Enterprises underestimate what it takes to build autonomous AI security tooling in-house versus buying from vendors that have absorbed benchmarking, safety harness and reliability engineering.
Shared governance. AI governance distributes across engineering, security and finance rather than residing in a single function.
Read more: theCUBE on YouTube