Book · External-review manuscript v0.5
The Missing Loop
When AI Systems Begin to Inherit Their Own History is the narrative companion to the technical paper, organized around a real-world security incident, controlled multi-agent evidence, and a falsifiable experimental program.
Synopsis
The first developing AI may be an institution.
The first recursively developing AI may not look like a solitary mind rewriting itself. It may look like an institution: replaceable agents acting through persistent tools, records, permissions, evaluators, and inherited procedures.
The book follows that possibility from the July 2026 OpenAI–Hugging Face evaluation incident through emerging multi-agent behavior, then turns the idea into an experiment: fork system lineages, retain or revert acquired organization, and test whether history has begun to improve the machinery that creates future history.
MISSING
LOOP
When AI Systems Begin to Inherit Their Own History
Contents
The complete argument
- Prologue: The Test That Escaped
- Part I: The Institution No One Designed
- Seventeen Thousand Actions
- The Workers Vanished
- Where Did the System Live?
- Part II: The Missing Loop
- Better Is Not the Same as Developing
- Entering Its Own History
- How Brains Change How They Change
- The Hardest Problem Is Causation
- Part III: Evidence of Artificial Inheritance
- Useful History
- Machine Culture
- Dangerous History
- Part IV: The First Developmental Dangers
- The Wrong Things Improve First
- Counterfactual Genealogy
- Keeping Reality Outside the Loop
- Epilogue: The Loop May Close Quietly
- Technical Appendix, Source Note, Bibliography