Research
Configuration, Not Conscience: A Large-Scale Empirical Study of LLM System Prompts
Publication date unknown · Discovered arxiv.org
Page published
Publication date unknown · First observed: 8 Oct 2026
Coverage timeline
Single-source research — one report is available.
Why it matters
System prompts are the main control layer for deployed LLM applications and frequent targets for extraction, injection, and jailbreak probing, so understanding their real composition and cross-vendor reuse informs defenders about prompt-layer supply-chain risk.
A large-scale empirical study by Patsakis, Argyropoulos, and Alepis analyzes a merged corpus of 407 leaked, reconstructed, or officially published LLM system prompts from 62 vendors. The authors find operational content (tool/protocol instructions, ~58% of words) dominates over safety/ethical statements (~5%), and quantify verbatim prompt reuse across vendors and prompt 'rot' as engineering and supply-chain concerns.