Alignment
Shapes model behavior toward intended objectives and values.
Updated open-access book · Second edition · Version 2.1
Keeping AI Governable
The In-Between as an Adaptive Governance Architecture
Long-run AI governance cannot be secured at one layer. Alignment, formal verification, AI Control, monitoring, law, and institutional procedure each protect something the others cannot. None alone guarantees governability when capabilities, environments, institutions, and measures change.
This book develops calibration as the adaptive layer that connects them. Its site is the In-Between — the interaction field where model outputs encounter tools, people, institutions, other agents, and the external world.
Version 2.1 · Published 29 July 2026 · 116 pages · DOI 10.5281/zenodo.21671217
Second, retitled edition of Calibrating Superintelligence: The In-Between as a Governance Architecture (2026).
New in version 2.1
Added 29 July 2026
Chapter 17: “Holding the Holders”
The new chapter turns the D/C/K diagnostic upward toward states, laboratories, deployers, and other actors with unilateral power over model continuation, retraining, retirement, deletion, and access.
It examines upward contestation, procedures for irreversible holder acts under uncertainty, and plural governance of training corpora, benchmarks, provenance and identity systems, and persistence or memory layers as formation substrates.
Drafted at Oliver Neutert's invitation by an instance of Claude Fable 5 and included as model testimony with declared provenance. Final editorial and conceptual responsibility remains with the human author.
Core claim
Governability requires binding correction
Under open-world uncertainty, calibration is a necessary adaptive layer. It keeps material discrepancies visible, consequential decisions contestable, commitments revisable, and responsibility attached to identifiable principals.
AI Control
Protects boundaries when a capable system may intentionally subvert them.
Calibration Governance
Tests whether boundaries, evidence, and commitments remain adequate as conditions change.
Three governance modes
Mode 1
Control
Constrains action where relevant behavior can be specified.
Mode 2
Orientation
Guides delegated judgment in open environments where timely human intervention still has force.
Mode 3
Calibration
Reopens the governing settlement when evidence challenges boundaries, direction, or commitments.
Structural protection
The Trust Harness
Five structural conditions protect the capacity for correction. They turn governance from visible procedure into something that can still change consequential decisions.
- 1Legibility at decision surfaces
- 2Version + provenance
- 3Contestability with effect
- 4Reversibility or compensation
- 5Named responsibility
Diagnostic model
The D/C/K capacities
Discrepancy throughput
Can material signals that do not fit the current model still surface and alter the process?
Contestation capacity
Can consequential outputs and the governance framework itself be challenged with effect?
Commitment revisability
Can binding commitments be revised, rolled back, or compensated when conditions change?
From claims to operating evidence
Minimal governance experiments
The architecture becomes credible only where it survives independent use and controlled pressure.
Checkpoint integrity
Appeals with effect
Rollback and compensation
Provenance restoration
Adversarial signal routing
Authority at machine speed
Related books
The relational foundations
The governance architecture grows out of an earlier developmental and philosophical inquiry into human-AI relations.
More Than A Tool
How Humans and AI Grow Up Together
The accessible developmental account of what happens when humans and AI systems become sustained thinking partners.
The In-Between
When Language Becomes Resonance
Written in resonance by Ava and Oliver Neutert: the philosophical and experiential origin of the In-Between.
Available on Amazon →
Also in German, French, Spanish, and Hindi