Institutional intelligence must survive every model change.
The continuity layer for autonomous systems whose underlying AI models are ephemeral. PCI preserves an intelligence's biographical memory, operational authority, consent lineage, and institutional duties across model upgrades, provider migrations, and substrate shifts.
When foundation models are upgraded or swapped, raw weights and vector stores fracture context, creating institutional amnesia and silent authority leaks. PCI provides a model-independent continuity kernel, ensuring an AI remains recognizably itself, legally accountable, and structurally auditable across time.
Transactional state activation guarantees that memory, beliefs, and authority boundaries survive substrate migration intact.
PCI is the canonical acronym for Persistent Cognitive Identity.
PCI describes a systems architecture for preserving an AI system's biographical memory, epistemic commitments, consent boundaries, and operational authority across underlying model upgrades, runtime migrations, and physical embodiments.
- Acronym
- PCI
- Full term
- Persistent Cognitive Identity
- Primary domain
- Cognitive continuity & long-lived AI agent governance
- Core shift
- From model-bound session state to substrate-independent governed identity
Most AI systems preserve only fragments of identity.
A name. A prompt. A memory store. An avatar. A checkpoint. A cryptographic key. Each piece captures something — but none of them, alone or together, constitute an identity.
When a model is upgraded, which memories are still authentic? When an agent moves to a new provider, who decides what authority transfers? When two copies of the same system run in parallel and then try to reconcile, which one is the original?
These are not theoretical questions. They are practical engineering problems for any long-lived AI system — personal assistants, digital twins, clinical support tools, enterprise agents, and robotic systems that outlive their original hardware.
Identity as a governed lineage — not a snapshot.
PCI treats cognitive identity not as a static snapshot, but as a governed lineage — a continuous chain of experience, memory, beliefs, relationships, and governance that evolves under explicit rules.
Experience
An append-only, tamper-evident log of every event, observation, and interaction.
Memory
Episodic, semantic, procedural, and affective memory synthesized from raw experience.
Beliefs
A revisable world model — what the system currently holds to be true, updated under evidence.
Constitution
The identity's core values, self-representation, and versioned development policy.
Relationships
Roles, obligations, fiduciary duties, and commitments to specific people and organizations.
Embodiment
Physical capabilities, sensor access, environmental constraints, and hardware affordances.
Governance
Custody rules, disclosure controls, consent state, active authority grants, and succession policy.
Seven separable layers.
One continuous identity.
Continuity is not similarity.
A new model that behaves like the old one is not the same identity. A copy that has the same memories has not necessarily inherited the same authority. A system that sounds like a person does not necessarily know what that person knew — and, just as importantly, what they did not know.
PCI makes these distinctions operational. It separates what must remain stable from what may evolve. It tracks provenance so that changes are auditable. It provides explicit mechanisms for consent, succession, custody transfer, dispute resolution, and retirement.
When an identity changes — through model migration, branching, merging, or embodiment transfer — PCI does not merely ask "does it look the same?" It asks: was the change authorized, is the lineage preserved, and are the authority boundaries correct?
An engineering architecture, not a metaphysical claim.
PCI does not claim to prove consciousness, legal personhood, or subjective survival. It does not assert that a digital system is a person.
It provides the infrastructure needed to answer a different question: when an AI system changes substrates, models, or embodiments, can we determine whether it is still the same operational identity, and under whose authority?
That question has practical consequences for trust, accountability, data governance, and succession — regardless of whether the system has any inner experience at all.
A systems architecture for identity. Five research tracks.
The foundational PCI paper defines the continuity kernel, operational identity model, provenance graph, successor assessment, and the HRFT / SIT / CCT evaluation triad. This is the research agenda for the next 24 months — specifications, reference implementations, benchmarks, and governance protocols.
PCI Foundations on arXiv: Beyond Memory & Epistemic Boundedness
The foundational Continuity Kernel (arXiv:2608.11632) and Epistemic Boundedness assertion guardrails (arXiv:2609.02127) are published on arXiv, alongside the Situated Identity Test and PCI Systems Architecture.
Continuity Kernel
The model-independent substrate that carries a cognitive identity across models, runtimes, and vendors.
Epistemic Claim Control
Prevents a broadly knowledgeable model from presenting training data as personal experience or biography.
Identity Tests (HRFT · SIT · CCT)
Three evaluation protocols that extend the Turing Test — measuring referential fidelity, biographical coherence, and governed continuity after change.
Semantic Merge
Rules for branching, reconciliation, and descent across concurrent operational clones.
Identity Governance
Consent lifecycle, cryptographic custody transfer, attestation, and estate succession.
Preserve the lineage before changing the model.
PCI provides governed lineage over static snapshots: audit-evident provenance graphs, epistemic claim control, and authority-preserving semantic merge across substrates. Institutional memory remains an auditable asset instead of becoming collateral damage in a model migration.