Instruction Bleed: Cross-Module Interference in Prompt-Composed Agentic Systems
arXiv:2606.26356v1 Announce Type: new Abstract: Practitioners of prompt-composed agentic systems report a recurring failure mode: editing one prompt module silently shifts the behavior of others despite no shared variable or executable dependency. We formalize this as compositional behavioral leakage (CBL): interference between modules sharing a context window. CBL is enabled by architectural non-isolation: transformer self-attention provides no formal boundary between concatenated modules. We p
Opening excerpt (first ~120 words) tap to expand
Computer Science > Artificial Intelligence arXiv:2606.26356 (cs) [Submitted on 24 Jun 2026] Title:Instruction Bleed: Cross-Module Interference in Prompt-Composed Agentic Systems Authors:Ching-Yu Lin, Yifan Liu View a PDF of the paper titled Instruction Bleed: Cross-Module Interference in Prompt-Composed Agentic Systems, by Ching-Yu Lin and 1 other authors View PDF HTML (experimental) Abstract:Practitioners of prompt-composed agentic systems report a recurring failure mode: editing one prompt module silently shifts the behavior of others despite no shared variable or executable dependency. We formalize this as compositional behavioral leakage (CBL): interference between modules sharing a context window.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv.org.
Discussion
0 commentsMore from arXiv.org
