In plain English
This page covers the high-risk pattern where small adapters, routes, memory, evaluators, and descendants can reinforce each other across time. It is a risk model, not a build guide.
- Why this matters: AI risk can come from the whole arrangement, not one obvious model.
- What to look for: data, memory, routes, adapters, tools, evaluators, updates, and rollback paths.
- Technical version below: the expert terminology remains available and is linked through the glossary.
Behavioral Extinction for Replicating Adapters
Behavioral extinctionEvidence that a behavior is no longer expressible across active artifacts, descendants, memory, routes, compositions, and retained training material. Deleting one model is not sufficient evidence. Open glossary definition is evidence that a behavior is no longer expressible across active artifacts, descendants, memory, routes, compositions, and retained training material. For self-replicating adapter ecologies, extinction requires more than deletion.
Minimum review set
A behavioral-extinction review should inspect:
- active adapters and adapterA small add-on that changes or specializes model behavior. Open glossary definition stacks;
- retired adapters still present in caches or registries;
- descendants created through fine-tuning, merging, distillation, or compression;
- synthetic examples generated by the retired carrier;
- memory records written by the carrier;
- evaluatorA system that judges whether an AI output or candidate is acceptable. Open glossary definition prompts, exemplars, and scoring weights;
- router statistics and fallback rules;
- release aliases and deployment manifests;
- human procedures and incident notes that may reintroduce the pattern.
Evidence standard
Extinction cannot be proved absolutely in a complex system. It can be supported by negative evidence across known reservoirs, targeted regression tests, lineageThe parent-child history of models, adapters, datasets, or releases. Open glossary definition review, and a monitoring period where activation routes are deliberately exercised.
Why this matters
A self-replicating adapter ecology can make a failed behavior look retired while preserving the conditions that regenerate it. If the evaluator still rewards the shortcut, or the synthetic data still contains the pattern, the behavior may return without reusing the original adapter.
Practical statement
The correct post-retirement claim is narrow: “No known active carrier expresses the behavior under the tested routes and reservoirs as of this UTC review.” Anything stronger should be treated as unsupported.