---
title: "Ilia Shumailov"
type: "entity"
entity_kind: "person"
status: "hub"
canonical_name: "Ilia Shumailov"
aliases: []
first_seen: "2026-09-14T00:00:00.000Z"
writer_model: "claude-sonnet-5"
connects_to: ["model collapse","bias amplification","recursive training","synthetic data","Nature (journal)"]
seek_code_commit: "546fa57"
---


Machine-learning security researcher (Oxford, later Google DeepMind) and
first author of "AI models collapse when trained on recursively generated
data" (*Nature*, 2024; preprint title "The Curse of Recursion"), the paper
that named and empirically demonstrated model collapse.

Matters to this vault as the methodological ancestor of two separate
claim clusters already resting on his work without, until now, a page of
his own: the general model-collapse literature
([[claim-model-collapse-recursive-training-erases-distribution-tails]])
and [[claim-wang-2024-bias-amplification-persists-independent-of-model-collapse|Wang
et al.'s bias-amplification experiment]], which explicitly builds its
ten-generation iterated-fine-tuning design "following Shumailov et al.
(2024)."

## References
- [[claim-model-collapse-recursive-training-erases-distribution-tails]]
- [[claim-wang-2024-bias-amplification-persists-independent-of-model-collapse]]
- Capture: 10-inbox/raw/2026-09-14-has-any-study-run-a-genuine-multi-generation.md