Expert perspectives on AI agent security, intent enforcement, and the new control plane for autonomous systems.
/ 001The bigger opportunity begins when intentd provides current objective state to the infrastructure running agents.
/ 002Response to OpenAI's call for collective cyber defense gets the urgency right.
/ 003The kernel enforces it now
/ 004The AI safety debate is focused on how powerful AI will become and whether we can keep increasingly capable models aligned. As agents act in the world, another question is becoming unavoidable
/ 005Unreliable generative AI should not be trusted with increasing power over the world. We agree and there is another path, we are building it.
/ 006Part 4 of our Intent Container series: enough architecture. Here is what our first reference implementation proves today and what it takes to make restore an authorization boundary.
/ 007AI agents can chain harmless-looking steps into something nobody approved.
/ 008When AI moves from answering citizens to representing them, legal boundaries cannot just be another instruction to the model.
/ 009Part 3 of our Intent Container series: suspend and resume exposes a security problem that process identity, workload identity, and sandbox isolation cannot solve on their own.
