Bounded Autonomy and Verifiable Safety for Agentic AI Enabled Automation

arXiv cs.LG Papers
agentic-ai safety bounded-autonomy formal-methods governance lyapunov simulation

Summary

该论文提出 BRaVeS（DNRS）框架，通过不变锚点、深度感知注意力与 Lyapunov 共识机制约束智能体推理的自主性，防止认知漂移，并在工业控制仿真中验证了有限步收敛与零安全违规。

arXiv:2610.08815v1 Announce Type: new Abstract: Agentic AI-enabled automation cannot be safely deployed in high-stakes environments on probabilistic reasoning alone. A recurring risk is epistemic drift: as reasoning deepens, system behavior may move away from subject-matter-expert constraints for safe operation. This paper presents BRaVeS, a bounded reasoning and safety-governance framework termed the Defensible Next-Gen Reasoning System (DNRS). BRaVeS encodes SME-defined constraints as invariant anchors, proposes MoDA-Style (Mixture of Depths Attention) depth-aware access as a candidate mechanism for keeping these anchors visible during inference, and uses a state hierarchy (SMARtAutonomy) to reduce autonomy as epistemic risk increases. To formalize bounded recovery, we introduce the Lyapunov-Bounded Consensus Framework (LBCF), which maps continuous epistemic-risk signals into a finite K-bag abstraction and applies shielded state transitions that enforce Lyapunov-style energy descent or route the system to a human-mediated terminal state. The formal convergence result applies to the finite LBCF abstraction under fixed thresholds and feasible-shield assumptions; it does not prove safety of the full continuous neural activation space. We evaluate the framework through a discrete event Monte Carlo simulation using HAI 22.04 industrial-control-system time-series data with synthetic noise and sensor-degradation regimes. Across the tested parameter-grouping strategies and thresholds, the LBCF process achieved finite-step convergence and no safety-guard violations. These results provide simulation-based evidence that bounded governance behavior can be enforced under the stated abstraction, while motivating future work on deployed transformer implementations, live human-in-the-loop validation, and broader adversarial settings.
Original Article
View Cached Full Text

Cached at: 10/08/26, 12:48 PM 

# Bounded Autonomy and Verifiable Safety for Agentic AI Enabled Automation
Source: [https://arxiv.org/abs/2610.08815](https://arxiv.org/abs/2610.08815)
[View PDF](https://arxiv.org/pdf/2610.08815)

> Abstract:Agentic AI\-enabled automation cannot be safely deployed in high\-stakes environments on probabilistic reasoning alone\. A recurring risk is epistemic drift: as reasoning deepens, system behavior may move away from subject\-matter\-expert constraints for safe operation\. This paper presents BRaVeS, a bounded reasoning and safety\-governance framework termed the Defensible Next\-Gen Reasoning System \(DNRS\)\. BRaVeS encodes SME\-defined constraints as invariant anchors, proposes MoDA\-Style \(Mixture of Depths Attention\) depth\-aware access as a candidate mechanism for keeping these anchors visible during inference, and uses a state hierarchy \(SMARtAutonomy\) to reduce autonomy as epistemic risk increases\. To formalize bounded recovery, we introduce the Lyapunov\-Bounded Consensus Framework \(LBCF\), which maps continuous epistemic\-risk signals into a finite K\-bag abstraction and applies shielded state transitions that enforce Lyapunov\-style energy descent or route the system to a human\-mediated terminal state\. The formal convergence result applies to the finite LBCF abstraction under fixed thresholds and feasible\-shield assumptions; it does not prove safety of the full continuous neural activation space\. We evaluate the framework through a discrete event Monte Carlo simulation using HAI 22\.04 industrial\-control\-system time\-series data with synthetic noise and sensor\-degradation regimes\. Across the tested parameter\-grouping strategies and thresholds, the LBCF process achieved finite\-step convergence and no safety\-guard violations\. These results provide simulation\-based evidence that bounded governance behavior can be enforced under the stated abstraction, while motivating future work on deployed transformer implementations, live human\-in\-the\-loop validation, and broader adversarial settings\.

## Submission history

From: Srini Ramaswamy \[[view email](https://arxiv.org/show-email/e60c8e67/2610.08815)\] **\[v1\]**Thu, 24 Sep 2026 02:09:49 UTC \(1,577 KB\)

Similar Articles

Verification and Self-Improvement in Agentic AI: Foundations and Limits

arXiv cs.AI

该论文提出一个理论框架，通过带隐藏终端随机性的有界验证来分析搜索、支持与检查器变化如何影响 Agentic AI 系统可接受的任务集合，并证明了独立多数放大、随机化验证器类的复杂度包含关系，以及在统一有界自修改下递归自我改进仍停留在同一验证类内的结论。

Evaluating Bounded Autonomy in Regulated Agentic AI: A Diagnostic Harness with Constitutional Rewards, Escalation Labels, and Runtime Governance

Hugging Face Daily Papers

The paper proposes RegLLM, a diagnostic harness for evaluating bounded autonomy in regulated agentic AI workflows, instrumenting six trustworthiness signals with constitutional rewards, escalation labels, and runtime governance. Small-scale Qwen2.5-3B pilots reveal that configuration variance can overwhelm apparent tuning effects on bounded-autonomy metrics, motivating larger evaluation sets and repeated runs.

Verifier-Grounded Agentic RL for Provably Correct Code

arXiv cs.LG

提出一种以程序验证器为奖励信号的智能体强化学习框架，通过读取验证器诊断信息（而非二元判决）并加入完整性审计，显著提升Qwen3系列模型在DafnyBench和SWE-Proof等验证基准上的正确率（最高达100%）。

Managed Autonomy at Runtime: Gear-Based Safety and Governance for Single- and Multi-Agent Cyber-Physical Systems

arXiv cs.AI

This paper presents EntropyRuntime, a discrete-time control system for single and multi-agent LLM-driven and robotic agents that uses five execution gears with utility-gated dispatch and event-driven fallback to ensure safety, stability, and continuity. It provides formal proofs and evaluates on a three-agent UR5 robotic assembly cell, achieving 99.6% anomaly detection rate.