Thank you for sharing this. I genuinely enjoyed reading it because it completely changed how I was thinking about AI security.

Before this, I tended to picture AI attacks through the traditional “hacker in a hoodie” lens. Your discussion of containment shifted my thinking toward a much more resilient question: not “Can we stop every attack?” but “What authority remains if an attack succeeds?” That was a subtle but important shift for me.

It also made me wonder about something a little further upstream. We probably can’t rehearse every possible attack, and there will always be unknown unknowns. Perhaps the goal isn’t predicting every attack, but identifying the incentives that are most likely to concentrate effort.

If incentives are among the strongest drivers of behavior, maybe they can also help prioritize where containment deserves the tightest tolerances. Rather than attempting to protect everything equally, organizations might focus first on the objectives that are most likely to attract concentrated effort—authority, sensitive information, critical infrastructure, financial assets, or influence.

That reminded me of manufacturing. We don’t inspect every possible failure equally. We establish tolerances around the components whose failure would have the greatest impact. Maybe AI security works similarly. Containment then becomes less about building an impenetrable wall and more about designing a system that continues functioning even after compromise.

If the containment is designed around the highest-impact scenarios, recovering from more routine attacks may become considerably easier. In that sense, containment becomes more than a technical safeguard; it becomes part of an organization’s long-term resilience and even its budgeting decisions. A well-contained attack doesn’t just reduce damage—it also becomes valuable data for improving the system without allowing failure to cascade.

Thank you again for such a thought-provoking article. It genuinely expanded how I think about AI security, and I’m looking forward to reading more of your insightful work.😁