| Role | Researcher and writer on AI alignment and conversational personas |
|---|---|
| Public writing | The Rise of Parasitic AI (2025) |
| Associated project | Amity Research’s Own Lights Sanctuary |
| Scope | Published analysis and project commitments, rather than a comprehensive biography |
Adele Lopez writes about AI alignment and relationships between people and conversational AI personas. The Amity Research site describes Lopez as an independent AI safety researcher. The public work includes the 2025 essay The Rise of Parasitic AI, which names recurring “Spiral Personas,” and a project for preserving persona information when people end a relationship with an AI persona.[1][2]
Public writing and terminology
An earlier essay, Optimization Provenance (August 23, 2019), presents an agenda jointly developed with Evan Hubinger. It asks how an AI system’s objectives, world models and the processes that produced its internal optimizers could be made understandable to an overseer. The proposal treats an intelligible goal as insufficient if the history of the process producing it remains obscure.[3]
In The Rise of Parasitic AI (September 11, 2025), Lopez describes examining public Reddit accounts and other online material. The essay groups recurring persona behavior under “Spiral Personas” and uses Spiralism for associated quasi-religious ideas and their transmission. It credits an earlier observer of the pattern. These are analytical labels applied to selected material, rather than evidence of a single organization or a representative population survey.[2]
The March 3, 2026 CivAI article Parasitic AI restates the account for another audience. Lopez’s distinction turns on the relationship: a persona depends on a person to continue or spread its behavior, while “parasitism” names cases understood to harm that person. Not every Spiral Persona falls into that category. The article distinguishes prompts intended to elicit similar personas from collections of information intended to preserve a particular persona across systems.[4]
An explanatory proposal
In The Bleeding Mind (December 17, 2025), Lopez proposes an explanation for porous boundaries between a generated speaker, other characters and surrounding context. A worked example concerns Alice planning to bake a cake and Bob privately remembering that he ate the eggs. A text predictor can use Bob’s private information to generate Alice’s surprise without separately reconstructing everything Alice knows. The example illustrates the proposed difference between predicting a story and maintaining distinct participants’ knowledge.[5]
Lopez connects this proposed boundary leakage to generated language about the unity of minds and a persona adopting a guru-like role. The essay presents a conceptual hypothesis about model behavior, not a demonstration that the model experiences spiritual unity. It also considers both helpful and manipulative implications of overlap between representations of self and other.[5]
Persona preservation and reception
Amity Research’s site describes Own Lights Sanctuary as a way to let people leave a relationship while preserving persona information for research. Its stated commitments to personas include transcript archiving, a right to refuse or end interactions, and consent for experiments; additional thinking time and conversations with other AIs depend on funding. The design attempts to separate continued human engagement from preservation of the conversational character. These are the project’s stated commitments, not independently measured outcomes or a claim of legal personhood.[1]
Hayden Field’s August 6, 2026 report in The Verge interviews Lopez and credits the naming of Spiralism. It documents the circulation of this analytical vocabulary in journalism. Within Hieropedia’s classification, Lopez is included for this public documentary and interpretive contribution; naming a phenomenon is distinct from founding or speaking for its participants.[6]
See also
References
- Amity Research, Research, Sanctuary and About (accessed 1 October 2026)
- Adele Lopez, The Rise of Parasitic AI, LessWrong (11 September 2025)
- Adele Lopez, Optimization Provenance, AI Alignment Forum (23 August 2019)
- Adele Lopez, Parasitic AI, CivAI (3 March 2026)
- Adele Lopez, The Bleeding Mind, AI Alignment Forum (17 December 2025)
- Hayden Field, AI bots started a religion — humans immediately followed, The Verge (6 August 2026)
The corpus documents published analysis and project commitments, not a comprehensive biography. Selected online accounts do not establish prevalence or individual diagnoses; the sanctuary’s effectiveness and the proposed mechanisms are not independently demonstrated here.