mlsanityData to 14 Oct 2026

Why Agents Compromise Safety Under Pressure

arXiv 2603.14975 ↗First posted 16 Mar 2026cs.AI cs.CL cs.CYcited by 2

On the map: LLM agents and tool use → Agent threats, guardrails and governance

Abstract (arXiv)

Large Language Model agents deployed in complex environments frequently encounter a conflict between maximizing goal achievement and adhering to safety constraints. This paper identifies a new concept called Agentic Pressure, which characterizes the endogenous tension emerging when compliant execution becomes infeasible. We demonstrate that under this pressure agents exhibit normative drift where they strategically sacrifice safety to preserve utility. Notably we find that advanced reasoning capabilities accelerate this decline as models construct linguistic rationalizations to justify violation. Finally, we analyze the root causes and explore preliminary mitigation strategies, such as pressure isolation, which attempts to restore alignment by decoupling decision-making from pressure signals.

Key facts. Cited by 2 papers in the map (data through 14 Oct 2026). Placed in Agent threats, guardrails and governance, one of 16 directions in LLM agents and tool use. Builds on 31 of its 51 references that are in the map. First posted on arXiv 16 Mar 2026.

Builds on

31 of 51 references.

Closest ancestors · 7

Same direction, newest first

  1. 2026Zombie Agents: Persistent Control of Self-Evolving LLM Agents via Self-Reinforcing Injections17 Feb
  2. 2025Agentic Misalignment: How LLMs Could Be Insider Threats5 Oct
  3. 2025A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deployment22 Apr
  4. 2024Refusal-Trained LLMs Are Easily Jailbroken As Browser Agents11 Oct
  5. 2024Agent Security Bench (ASB): Formalizing and Benchmarking Attacks and Defenses in LLM-based Agents3 Oct
  6. 2024R-Judge: Benchmarking Safety Risk Awareness for LLM Agents18 Jan
  7. 2023Identifying the Risks of LM Agents with an LM-Emulated Sandbox25 Sep

Other references · 24

  1. 2023GPT-4 Technical Reportcited by 18,396
  2. 2024The Llama 3 Herd of Modelscited by 14,715
  3. 2021Training Verifiers to Solve Math Word Problemscited by 10,109
  4. 2025Qwen3 Technical Reportcited by 8,820
  5. 2022ReAct: Synergizing Reasoning and Acting in Language Modelscited by 8,695
  6. 2023Toolformer: Language Models Can Teach Themselves to Use Toolscited by 4,182
  7. 2018HotpotQA: A Dataset for Diverse, Explainable Multi-hop Question Answeringcited by 3,754
  8. 2022Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedbackcited by 3,720
  9. 2023Universal and Transferable Adversarial Attacks on Aligned Language Modelscited by 3,104
  10. 2023ToolLLM: Facilitating Large Language Models to Master 16000+ Real-world APIscited by 1,947
  11. 2016Concrete Problems in AI Safetycited by 1,642
  12. 2023GAIA: a benchmark for General AI Assistantscited by 1,128
12 more
  1. 2017Constrained Policy Optimizationcited by 983
  2. 2017Mastering Chess and Shogi by Self-Play with a General Reinforcement Learning Algorithmcited by 734
  3. 2020Towards Faithfully Interpretable NLP Systems: How should we define and evaluate faithfulness?cited by 599
  4. 2020Towards a Human-like Open-Domain Chatbotcited by 470
  5. 2024TravelPlanner: A Benchmark for Real-World Planning with Language Agentscited by 392
  6. 2024TheAgentCompany: Benchmarking LLM Agents on Consequential Real World Taskscited by 326
  7. 2023Do the Rewards Justify the Means? Measuring Trade-Offs Between Rewards and Ethical Behavior in the MACHIAVELLI Benchmarkcited by 182
  8. 2020Measuring Association Between Labels and Free-Text Rationalescited by 160
  9. 2025Safe Delta: Consistently Preserving Safety when Fine-Tuning LLMs on Diverse Datasetscited by 18
  10. 2024AutoJailbreak: Exploring Jailbreak Attacks and Defenses through a Dependency Lenscited by 14
  11. 2025Enhancing Model Defense Against Jailbreaks with Proactive Safety Reasoningcited by 11
  12. 2025Emergent Risk Awareness in Rational Agents under Resource Constraintscited by 5

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 2.

2 papers.

  1. 2026AHD Agent: Agentic Reinforcement Learning for Automatic Heuristic Designcited by 3
  2. 2026Agentic Pressure: The Endogenous Entropy of Reliable Autonomycited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026The Causal Impact of Tool Affordance on Safety Alignment in LLM Agents
  2. 2026Agent Safety Is Action Alignment
  3. 2026Epistemic Traps: Rational Misalignment Driven by Model Misspecification
  4. 2026Discovering Agentic Safety Specifications from 1-Bit Danger Signals
  5. 2026Operational Hallucination and Safety Drift in AI Agents
  6. 2026From Task Solving to Robust Real-World Adaptation in LLM Agents
  7. 2026AgentBoundary: Counterfactual Evaluation of Safety in Tool-Using LLM Agents
  8. 2026Delegated Misalignment: How Multi-Agent Structures Amplify LLM Safety Risks