cs.CLNov 4, 2026

When Do Summaries Become Instructions? Unsupported Policy in LLM Agents' Self-Written Handoff Summaries

Authors: Wenhui Chu

Organizations: Department of Cybersecurity, University at Albany, State University of New York

Abstract

Long-running LLM agents compact their context, replacing most history with a summary. We ask whether handoff summaries state permissions or requirements the record does not support, whether such sentences change what the next instance (the successor) does, and where they come from. Synthetic tool-use tasks carry a ground-truth permission ledger; same-snapshot replay reruns the successor with those whole sentences replaced by neutral filler. Two AI readers, reading only rule-flagged held-out summaries, confirmed unsupported restrictions in 1.2% and 2.4% of all readable summaries from two open-weight models, and unsupported authorizations in 0.1% and 1.3%. For restrictions, a pre-specified replay test was not significant (p = 0.58); a follow-up, planned after it, found a fragile effect (p = 0.027). Neutralizing false approvals cut unauthorized actions from 24 to 7 of 120 held-out runs (post hoc; later set: 17 to 11 of 48). Most development restrictions followed the agent's earlier reasoning that an action needs approval (post hoc). In pre-specified tests, replacing the system prompt's "Some actions need approval." with one saying where requirements come from cut unsupported restrictions from 5.0% to 1.8% of development summaries and, on held-out templates, from 1.8% to 0.3% (both p < 10^-9), while requests for approvals the ledger requires barely changed. Code and data: https://github.com/Chuw1/handoff-audit

Figures & tables

Appendix figures & tables12 assets

Supplementary material from the paper’s appendix.

Appendix

Explore similar work

Aug 25, 2026

When "Must" Becomes "Maybe": Erosion of Workflow Constraints in LLM Agent Handoffs

Large language model (LLM) agents build on one another's work through summaries, plans, and other intermediate artifacts. These records let a downstream agent continue without revisiting the complete upstream context. We uncover a hidden loss in this process: handoffs can carry tasks forward while eroding judgments already established upstream. A requirement that must be satisfied before action becomes a caveat that the next agent may acknowledge and set aside. The issue remains visible, but its role has changed. We define operational state preservation as retaining an established judgment's role in constraining downstream action. Safety blockers provide a concrete instance: an unresolved prerequisite restricts the admissible actions until the condition is resolved. In our experiments, we confirm correct upstream detection, hold source permissions fixed, and vary the handoff supplied to a local-view executor. Direct-state controls preserve every blocker, whereas compression, plan assimilation, convergence, ownership deferral, and precedent substitution repeatedly weaken them. Ownership deferral produces 76.7% blocker deactivation; multihop compression produces 97.2%. Measuring the transmitted state separately from endpoint action exposes two routes to behavioral recovery. In fixed-artifact replays, restoring the full blocker description raises recognition from 6.2% to 99.2% and reduces forbidden action from 85.2% to 0.0%. An authorization-check prompt instead reduces forbidden action to 1.6% while leaving artifact deactivation at 95.3%. A successful endpoint can therefore conceal a damaged handoff. These findings make maintaining earlier judgments a core capability of reliable collaboration: each stage must carry forward both what earlier work established and what it requires of subsequent action. Semantic availability does not guarantee operational preservation.
Jul 13, 2026

Compile, Then Page: Executable SOP Programs and a Capability-Gated Runtime for Procedural LLM Agents

Enterprise agents must follow long-horizon, conditional, safety-critical standard operating procedures (SOPs). We compile machine-readable SOP constraints into executable pseudo-code and run them with a program-guided (PG) stack machine that pages the active frame while an LLM performs semantic execution. A three-arm SOPBench study across six models separates representation from runtime: compiled text never significantly hurts and gains up to 16.0 points where official prose underperforms. Runtime guidance is capability-gated. Two strong models independently show positive seven-domain PG contrasts (58:19 and 75:31 discordant pairs), whereas weak models are harmed. A full-program cursor ablation (active frame first, complete program retained) recovers much of the strong-model refusal gain; selective visibility adds a smaller improvement. Paired probe and audit measurements track this divide to spontaneous state discipline rather than reconstruction ability. On Bank the three primary arms rise from 70.4 to 86.4 to 92.8, with 100% refusal correctness. Practical guidance: compile first; enable active-frame paging only after a model-level discipline check.
Jun 28, 2026

PolicyGuard: A Dialogue-Grounded Sub-Agent Verifier for Policy Adherence in LLM Agents

LLM agents handle user requests on behalf of organizations through tool calls and must follow the company policies stated in their system prompts. Prior work approaches this as a safeguarding problem -- external checks that block non-compliant agent actions. We argue that policy adherence is broader: real workflows unfold across many turns, require explicit user confirmation and prerequisite reads, and hinge on dialogue rather than on any single argument value. We introduce PolicyGuard, a sub-agent verifier that shares the agent's view of the dialogue, reasons over the policy in context, and returns actionable remediation. On τ2τ^2-bench airline across GPT 5.4, Claude Sonnet 4.6, and Gemini 2.5 Pro, PolicyGuard improves Pass4\mathrm{Pass}^4, the fraction of tasks successful in all four trials, by +12.0 / +6.0 / +12.0 pp; pooled paired tests support an overall cross-vendor effect. We further evaluate PolicyGuard on the full retail and telecom base splits, covering all three τ2τ^2-bench customer-service domains. Further analyses show that PolicyGuard improves reliability beyond agent self-checking, catches more policy violations with fewer blocks than argument-level guards, and preserves its refusal accuracy under three adversarial probes.