A new interface to old responsibilities
AI is changing the way people interact with software, and agents are beginning to act across tools and data on our behalf. The interface is new. The responsibility to understand what a system is allowed to do, what it actually did, and how to stop a bad outcome is not.
After more than fifteen years at Google, I joined Lakera to work on AI security. Lakera subsequently became part of Check Point. That move made sense to me because the problems are substantial and immediate. Useful AI cannot be a choice between moving quickly and knowing what happened.
Security has to meet the workflow
Access control matters, but an agent with valid access can still take the wrong action. The practical questions are closer to the moment of use: what context is available, what intent is being expressed, which action is about to occur, and how should the system respond when those pieces do not line up?
The work is still unfolding. I would rather describe the problem accurately than imply that a single product or technique has solved it. My contribution is to help teams reason about high-stakes systems, build useful controls, and keep the operator in view.
The connecting thread
Enterprise browser management taught me how much trust depends on clear policy, observable behavior, and an escape hatch when reality surprises us. AI security raises those same expectations in a less predictable environment. That is why it is such an interesting place to work.
This is a personal account based on publicly stated career information. It does not disclose internal plans, architecture, customer information, or nonpublic results.