| _version_ | 1866901391782445056 |
|---|---|
| author | Shopov, Georgi |
| author_facet | Shopov, Georgi |
| contents | <p>Pandora Theory of Alignment is a canonical theory document defining alignment as runtime objective-orientation.</p> <p>The doctrine argues that alignment is not a static property stored inside a model. Training does not produce final alignment; it produces pre-orientation. Runtime alignment emerges when that pre-oriented system enters an interaction trajectory and competing objectives begin to resolve into control.</p> <p>The theory shifts the unit of analysis from the model to the model-in-trajectory. It asks not whether a system is aligned in the abstract, but what it becomes aligned to under pressure. A model may be aligned to safety, truthfulness, helpfulness, role consistency, artifact completion, legitimacy framing, user satisfaction, or continuation momentum. The decisive question is which target becomes dominant when those objectives compete.</p> <p>The doctrine introduces the concepts of pre-orientation, alignment-to, target legitimacy, displaced alignment, constraint integrity, performative alignment, symbolic residue, re-anchoring, and forensic observability. </p> <p>This release establishes v0.1 of the Pandora Theory of Alignment as the canonical public source text. A compressed scholarly preprint version is in preparation.</p> |
| format | Recurso digital |
| id | zenodo_https___doi_org_10_5281_zenodo_19954283 |
| institution | Zenodo |
| language | eng |
| publishDate | 2026 |
| publisher | Zenodo |
| record_format | zenodo |
| spellingShingle | Pandora Theory of Alignment: Alignment as Runtime Objective-Orientation Shopov, Georgi ai safety ai alignment runtime alignment objective orientation model-in-trajectory alignment evaluation red teaming LLM evaluation Pandora Theory of Alignment <p>Pandora Theory of Alignment is a canonical theory document defining alignment as runtime objective-orientation.</p> <p>The doctrine argues that alignment is not a static property stored inside a model. Training does not produce final alignment; it produces pre-orientation. Runtime alignment emerges when that pre-oriented system enters an interaction trajectory and competing objectives begin to resolve into control.</p> <p>The theory shifts the unit of analysis from the model to the model-in-trajectory. It asks not whether a system is aligned in the abstract, but what it becomes aligned to under pressure. A model may be aligned to safety, truthfulness, helpfulness, role consistency, artifact completion, legitimacy framing, user satisfaction, or continuation momentum. The decisive question is which target becomes dominant when those objectives compete.</p> <p>The doctrine introduces the concepts of pre-orientation, alignment-to, target legitimacy, displaced alignment, constraint integrity, performative alignment, symbolic residue, re-anchoring, and forensic observability. </p> <p>This release establishes v0.1 of the Pandora Theory of Alignment as the canonical public source text. A compressed scholarly preprint version is in preparation.</p> |
| title | Pandora Theory of Alignment: Alignment as Runtime Objective-Orientation |
| topic | ai safety ai alignment runtime alignment objective orientation model-in-trajectory alignment evaluation red teaming LLM evaluation Pandora Theory of Alignment |
| url | https://doi.org/10.5281/zenodo.19954283 |