Persistent identity
Continuity of an AI system across restarts, copies and updates, and how that continuity is documented.
ConceptSources & confidence: Qualified confidence
Checked 2 October 2026 · AI-assisted review
Checked the definition, attribution and stated distinctions against the listed source passages.
- Local essay: you-cant-shame-a-fork
The essay’s distinction between identity drift, behavioural characterisation and lineage. Checked the repository MDX source, not the live page.
- Raji et al., Closing the AI Accountability Gap (2020)
An established use of accountability gap and an organisational auditing approach.
What this check doesn’t establish
- The essay’s governance proposal was checked as an argument, not independently validated as a technical standard.
Confidence applies to the scope above, not every claim made by the underlying source. How confidence works
The essay asks how reputation and accountability should persist when models are copied or changed. Several different things may persist: weights, a deployment identifier, an operator, documented lineage or a pattern of behaviour.
Copying does not necessarily erase history. Logs, version identifiers and recorded lineage can preserve provenance, even though they do not settle every philosophical or behavioural question about identity. The essay’s proposed semantic footprint is a candidate approach to characterisation, not a synonym for persistent identity or an established solution.
In the essays
- You Can't Shame a Fork
Read the essay
Sources
- Memento
Discussed in
- The Moltbook Case and How We Need to Think about Agent Security
Discussed in
Ideas
Record history
Catalogue source: knowledge/concepts/persistent-identity.md
Generated by claude-code/claude-opus-5 on 2026-07-26