What the study found
The paper defines AI drift as authority-vacancy transition instability under the absence of final arbitration. It presents AI drift as a lower condition than alignment: a pre-alignment instability in which authority assignment, refusal, world-binding, and editability have not stabilized.
Why the authors say this matters
The authors suggest that alignment should be understood as drift governance without closure, meaning transition conditions should be designed so that coercive or false stabilizers do not become cheap defaults. They say refusal, world-binding, rollback, and authority editing should remain operationally available.
What the researchers tested
The paper uses the SΔϕ Formalism and builds from an earlier distinction between how humans and AI systems handle the absence of a final arbiter. It formalizes AI drift with minimal axioms, a compact operational schema, failure modes, and diagnostic tests, and it distinguishes several drift types.
What worked and what didn't
The paper distinguishes AI drift from data drift, hallucination, freedom, and misalignment. It identifies under-authority drift, borrowed-authority drift, world-binding drift, over-compliance drift, and over-authority closure, and it argues that AI drift names the encounter when transition continues without legitimate, editable, world-bound authority.
What to keep in mind
The abstract does not report empirical evaluation of current AI systems or performance results. It also states that the paper does not claim present AI systems possess phenomenological experience, existential self-knowledge, or belief.
Key points
- AI drift is defined as authority-vacancy transition instability under the absence of final arbitration.
- The paper treats AI drift as a pre-alignment instability, not the same as data drift, hallucination, freedom, or misalignment.
- The authors argue that alignment should be understood as drift governance without closure.
- The paper distinguishes under-authority drift, borrowed-authority drift, world-binding drift, over-compliance drift, and over-authority closure.
- The abstract says the paper does not claim AI systems have phenomenological experience, existential self-knowledge, or belief.
Disclosure
- Research title:
- AI drift is defined as authority-vacancy transition instability
- Authors:
- Sofience
- Publication date:
- 2026-04-28
- OpenAlex record:
- View
Get the weekly research newsletter
Stay current with peer-reviewed research without reading academic papers — one filtered digest, every Friday.