Every enterprise control attaches to an actor. Trust infrastructure answers, for every agent system: who is it, what is it allowed to do, what has it done, why did it act, and is it still the same one over time? Since 2017, Olbrain Labs has worked on the deepest of these questions — who the agent is, and whether it is still the same one — and on the verifiable record that rests on it.
Volvo is safety. Not because they market it — because every engineering decision they make flows from that constraint. Safety is not a feature they bolt on. It is the design principle that shapes the entire vehicle.
Olbrain is trust. Self-identity is the design constraint that makes it possible — the architectural decision that shapes every agent system we build, the way crumple zones shape every Volvo.
Without these three properties, there is no stable actor. Without a stable actor, there is no accountability. And without accountability, AI remains a tool you cannot rely on.
Three properties that define what it means for a machine to have identity. If any one is missing, identity breaks down. CNE is the engineering specification behind every identity we create — the measurable criteria that determine whether an agent has genuine identity or is merely simulating one.
An agent's behaviour, decisions, and responses remain internally consistent. Its actions align with its stated values, objectives, and constraints — not just in one conversation, but across every interaction it has ever had.
The agent maintains a continuous thread of experience. It remembers what it has done, what it has learned, and what commitments it has made. Context doesn't reset. History doesn't vanish. The story holds together.
Each agent identity is singular and non-duplicable — one un-copyable agent. It cannot be forked, cloned, or impersonated. Continuity alone cannot decide which ship is real; continuity plus an externally enforced rule that only one can hold the identity settles it. This is what makes accountability possible — a genuine actor that records can belong to, that trust can attach to.
Not every agent needs the same depth. Our architecture distinguishes between what the enterprise already has and what it is missing.
Authentication, access control, permissions — the enterprise’s existing identity infrastructure. External identity decides what an agent may access, which systems it may call, and under whose authority. Necessary, but not sufficient: it governs the door, not the actor behind it.
The deeper substrate: a continuous, verifiable actor that governance can attach to. Self-identity is what makes an agent attributable and auditable over time — the actor the record belongs to. Access control decides what an agent can touch. Self-identity establishes who it is.
Nine years of asking the same question from deeper and deeper angles.
The question that started it all: can machines have a coherent sense of self? We began exploring what it would take to build cognitive continuity into AI systems — not consciousness in the human sense, but a stable, continuous substrate that holds together over time.
The research narrowed from broad consciousness to something more precise: identity. What makes an agent the same agent across interactions? We started formalizing the properties that a machine identity would need — coherence, continuity, and exclusivity.
Coherence, Narrative Continuity, Exclusivity — the three pillars crystallized into a formal protocol. CNE became the technical framework for everything we build: the proof that an agent is who it says it is, across time.
The architecture matured into two distinct layers: external identity (IAM — the enterprise’s existing controls) and self-identity (the deeper substrate that makes an agent a verifiable actor). Self-identity makes external identity meaningful — the design constraint that shapes everything.
The research becomes infrastructure — trust infrastructure, with self-identity underneath. Olbrain ships as the Agent OS for enterprises: build, run, and govern the fleet. Every agent system is attributable and auditable by construction.
Research is never finished. These are the questions that drive our current work.
When hundreds of agents operate under one enterprise, how do individual identities compose into a coherent organizational intelligence? This is fleet-level governance — the hardest open question in enterprise AI.
Memory is a component of identity, but identity is more than memory. Where exactly does the line fall? What can an agent forget and still remain the same agent?
Current architectures require deliberate design of the identity substrate. Is there a path where coherent self-identity develops through interaction — and would we want that?
The accountability chain is clear in theory. In practice, how do enterprises audit, verify, and trust an agent’s identity claims? What does the proof layer — the signed record, the verifiable actor — actually look like at scale?
Within one enterprise, the fleet shares a trust root. Across enterprises — a supplier’s agent talking to a buyer’s agent — there is no shared root. How does cross-company trust work without a central authority?
If you are thinking about trust, identity, and what it means for machines to be accountable — we would like to hear from you. Whether you are a researcher, an enterprise building with agents, or someone who believes AI needs more than intelligence.