Home / Articles / System prompts as behaviour policies / Versions / v1.1.0
System prompts as behaviour policies: what changed in v1.1.0
From v1.0.0 (2026-10-14, a5d5f7d8e) to v1.1.0 (2026-10-27, db67c5ce1), paragraph by paragraph.
0 paragraphs added, 0 removed, 2 changed in place, 61 unchanged. About 0 words added and 87 removed. Insertions are marked like this, deletions like this; unchanged runs are folded to one line; figures appear as their file names.
2 unchanged paragraphs
!shot sp-map.webp | images/ | Claude Haiku 5.5's published system prompt as policy rows. 4,255 words, 181 sentences, 87 that read as rules and 82 once explanations are set aside,rows, sorted into eight kinds,kinds withby where each kind could be kept other than in the model's good behaviour: 36 rules by the harness, 21 in part by classifiers, 25 by the model alone.kept. | comparison
37 unchanged paragraphs, under In short {#in-short}, The hypothesis {#hypothesis}, The prompts {#prompts}…
!shot sp-ontology.webp | images/ | A system prompt as a policy file.file: Aa policy per layer, the vendor's, the deployer's and the user's, each with aits mandate, ainputs, reach, its inputsstate and how far each is trusted, the state that changes the rules mid-conversation, and its rules: each with a kind, a trigger, a decision, a reason, an authority, a barrier, what enforces it, tests and the source text it came from.rules. | architecture
22 unchanged paragraphs, under Is anyone using this for real? {#real}, What this does not claim {#limits}, Where this comes from {#sources}