Compaction can leave a safety rule textually present but behaviorally dead, so presence-based audits are insufficient; rule-form text also survives better than matched facts.
Canonical run: Qwen (20260602_214506, 1200-token output cap); Claude output-cap cell: 20260804_223216 (150-token cap, §4.5); same-design Llama run: 20260610_150139 (Appendix H.3)
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CR 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
AI Guardrail Survival under Single-Cycle Agentic Self-Summarization
Compaction can leave a safety rule textually present but behaviorally dead, so presence-based audits are insufficient; rule-form text also survives better than matched facts.