A technical framework, evidence review, and experimental proposal for distinguishing a better artifact from a better improver.
Recursive development · causal identification
THE MISSING LOOP
The system does not simply change. It changes how it changes.
A falsifiable hypothesis about the next phase of AI development: when systems enter their own causal history.
The hypothesis
Optimization, adaptation, self-modification, and genuinely recursive development are not the same.
Explore →Evidence & cases
What recent systems and events demonstrate, and what they still do not establish.
Explore →The experiment
The decisive retained-versus-reverted lineage test, controls, measures, and falsifiers.
Explore →Risks & boundaries
Causal error, evaluator capture, inherited faults, escalation, lineage drift, and rollback loss.
Explore →Recent evidence
Parts of the predicted substrate are appearing.
OPENAIHUGGING FACE
OpenAI–Hugging Face incident
Containment failure, durable external artifacts, and operation across short-lived workers. Suggestive of a broader system boundary, not proof of recursive leverage.
Case study →
Anthropic multi-agent research
Conflict, interference, causal misattribution, and in some runs negotiated repair inside a shared environment. No demonstrated cross-reset inheritance.
Case study →
Recursive harness research
Systems can modify parts of their own prompts, tools, evaluation scaffolds, and improvement procedures. The decisive causal fork remains open.
Case study →The first recursively developing AI may not look like a solitary mind. It may look like an institution whose workers vanish while its procedures and history remain.
What begins whenthe loop closes?
