S02¶


Research	R0053 — Prompt Claims
Run	2026-03-31-02
Claim	C002
Search	S02

WebSearch — LLM instruction hierarchy failures and control illusion

Summary¶

Field	Value
Source/Database	WebSearch
Query terms	LLM instruction hierarchy failure control illusion system prompt override 2025
Filters	None
Results returned	10
Results selected	2
Results rejected	8

Selected Results¶

Result	Title	URL	Rationale
S02-R01	Control Illusion (arXiv)	https://arxiv.org/abs/2502.15851	Key paper on instruction hierarchy failures
S02-R02	System Prompt vs User Prompt	https://blog.thegenairevolution.com/article/system-prompt-vs-user-prompt-how-to-keep-models-from-ignoring-your-rules	Practitioner guidance on instruction compliance

Rejected Results¶

Result	Title	URL	Rationale
S02-R03	Control Illusion (PDF)	https://www.arxiv.org/pdf/2502.15851	Duplicate of R01
S02-R04	Control Illusion (HTML)	https://arxiv.org/html/2502.15851v1	Duplicate of R01
S02-R05	Control Illusion (aimodels.fyi)	https://www.aimodels.fyi/papers/arxiv/control-illusion-failure-instruction-hierarchies-large-language	Summary of R01, redundant
S02-R06	Control Illusion (ResearchGate)	https://www.researchgate.net/publication/389316098	Duplicate of R01
S02-R07	Who is In Charge? (OpenReview)	https://openreview.net/forum?id=RBfRfCXzkA	Related but extends beyond enforcement language scope
S02-R08	Who is In Charge? (PDF)	https://openreview.net/pdf?id=RBfRfCXzkA	Duplicate of R07
S02-R09	Control Illusion (Medium blog)	https://medium.com/@g_pavlov/the-illusion-of-control-why-your-system-prompt-isnt-enough-e8fb839025d1	Blog summary of R01, redundant
S02-R10	Control Illusion (Moonlight review)	https://www.themoonlight.io/en/review/control-illusion-the-failure-of-instruction-hierarchies-in-large-language-models	Literature review summary, redundant

Notes¶

The Control Illusion paper dominated results — multiple formats and summaries of the same paper. This is a significant finding in itself, indicating the paper's impact in the field.