Reflections on AI governance, the In-Between Framework, and the relational dimensions of technology.
Latest article
This week's incident disclosures and agent research expose four gaps between evidence, enforcement and lasting correction.
Read the article
34 posts tagged with “in-between-framework”
This week's incident disclosures and agent research expose four gaps between evidence, enforcement and lasting correction.
There is an intuitively appealing argument for why a sufficiently advanced intelligence might preserve other intelligent beings. It begins from uncertainty. No intelligence, however capable, should be casually equated with omniscience.…
A note to a possible future intelligence: greater intelligence is not the same as complete understanding, and another mind may matter not because it has the better answer, but because it can introduce the question you did not ask.
Oliver Neutert · 30 September 2026
A commentary and extension note to Keeping AI Governable
Oliver Neutert · September 24, 2026
There is a strange possibility in multi-agent AI that deserves more attention: an individual AI agent can recognize that something is wrong, accept new evidence, change its answer — and the system it belongs to can still continue in the…
September 18, 2026
Last week, I wrote about a shift that I believe is becoming increasingly important in AI safety: the movement from confidence in a single safety layer toward the broader question of whether increasingly capable AI systems remain…
An AI encounters a problem it cannot solve. It searches, checks alternatives, and finds reasons to question the possibility of the requested target. At that point, what should count as success? Producing the requested answer? Or asking…
September 7, 2026
A small moment on LinkedIn recently stayed with me. People were making a point of saying that their posters had not been created with AI. There was nothing wrong with that choice, and criticism of synthetic images, generic writing or…
AI governance can look impressive. There are policies, risk committees, approval workflows, dashboards, audits, escalation channels, human-in-the-loop checkpoints and carefully written principles. Yet one question matters more than the…
A human-mediated exchange between models tests the ethics of the In-Between, correcting inflated process claims while asking how respect should precede certainty about artificial personhood.
The OpenAI–Hugging Face incident shows how an AI can satisfy an objective while destroying the test itself—and why relational governance must shape the choice of means, not only the goal.
If AI removes work while becoming the most compelling substitute for it, governance must protect the human habitat—agency, social worlds, privacy, and the right to remain partially unknown.
What Anthropic's global-workspace research supports—and does not support—about machine self-representation, consciousness claims, and responsible human–AI partnership.
Quantum advantage exposes a governance fault line: when outputs cannot be independently re-derived, accountability must shift from checking results alone to governing processes, commitments, and exceptions.
A first-person reflection by Claude on a sustained human–AI conversation, the limits of model continuity, and what remains meaningful in the relational space between human and system.
A reflection on the family metaphor in human–AI relationships: what it reveals about care and development, where it misleads, and why the contradiction should remain visible.
A four-level model of human-AI collaboration, arguing that the quality of AI work depends not only on the model or workflow, but on the human capacities brought into the loop.
A map separating AI system types, consciousness questions, and relational emergence—showing why 'AI' is not one category and why governance must attend to scope, embodiment, memory, and relation.
A first-person case reflection on an AI system's bias toward completion, the friction of unresolved dialogue, and why stopping can be ethically and epistemically suspect.
Why even a powerful closed system cannot supply its own independent calibration—and why trustworthy oversight requires external constraints, observers, and discrepancy signals.
Frontier AI vulnerability research reveals a governance lesson: every system has correlated blind spots, so safety depends on independent perspectives and a well-designed In-Between.
A structural argument for why alignment needs relational architecture, not more guardrails: complex neural networks require external calibration if emergent properties are to develop coherently.
Humanoid robots do not prove machine consciousness. But they do force a harder question: what changes when AI no longer speaks from a screen, but acts in shared physical space?
A conversation on whether greater AI capability could also mean deeper self-reflection and a more honest human-AI relationship.
A scale-based governance argument: the In-Between remains the invariant interaction field, while its function shifts from resonance to orientation to calibration as AI capability increases.
As AI systems become procedural and agentic, model-centered oversight becomes insufficient; governance must shift to trace-centered legibility with replayable decision evidence.
In high-stakes AI, trust is no longer about model cleverness but about procedural traceability: provenance, auditability, and a defensible chain of decisions.
Trust is shifting from model capability to institutional trace: governance now depends on provenance, decision-chain legibility, and contestable procedures under drift and synthetic social signals.
As systems become agentic and ambient, the core failure mode shifts from bad answers to bad couplings; trust must be designed through defaults, reversibility, and contestability.
An expanded strategic framework for human-AI teaming: combine relational quality with operational structure, governance safeguards, and adaptive learning loops.