AblitAI Technical Report on Agentic Self Replication
A recorded host-managed Garden lineage and a theoretical framework; autonomous cross-computer replication remains unproven.
THE RESEARCH INDEX
Findings and
open questions.
Experiments from our local research, alongside clearly labelled plans for future studies. Every finding includes its evidence and limits.
Browse findings and study plans
Testing a hardened corpus
A discovery pass found two unintended weaknesses in a synthetic 20-host corpus; both were fixed and retested.
When a security control gives false confidence
A lab control appeared to confirm a fix while checking the wrong result; later review exposed the mistake.
A PHP crash comparison
A local vulnerable-versus-patched comparison reproduced a crash associated with an already disclosed PHP vulnerability.
A small model-training screen
A four-case local screen found unchanged answers after a small training segment.
When a longer context stops being useful
A research question about memory, retrieval and answer quality. The setup, method and limitations belong alongside the result.
What fits on a consumer GPU?
Model size, memory use and the compromises behind a usable setup.
When coding clients expose model limits
What integration tests can teach us about tools, repetition and reliability.
No research matches those filters.
Try another phrase or clear the filters to see every topic.