mlsanityData to 7 Oct 2026

Safety is Non-Compositional: A Formal Framework for Capability-Based AI Systems

arXiv 2603.15973 ↗First posted 16 Mar 2026cs.AIcited by 4

Not on the map. Why?

Abstract (arXiv)

This paper contains the first formal proof that safety is non-compositional in the presence of conjunctive capability dependencies: two agents each individually inca- pable of reaching any forbidden capability can, when combined, collectively reach a forbidden goal through an emergent conjunctive dependency.

Key facts. Cited by 4 papers in the map (data through 7 Oct 2026). Builds on 5 of its 49 references that are in the map. First posted on arXiv 16 Mar 2026.

Builds on

5 of 49 references.

References · 5

  1. 2022ReAct: Synergizing Reasoning and Acting in Language Modelscited by 8,355
  2. 2023ToolLLM: Facilitating Large Language Models to Master 16000+ Real-world APIscited by 1,880
  3. 2023Gorilla: Large Language Model Connected with Massive APIscited by 1,302
  4. 2023AgentBench: Evaluating LLMs as Agentscited by 1,211
  5. 2023TaskBench: Benchmarking Large Language Models for Task Automationcited by 138

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2026: 4.

4 papers.

  1. 2026When Safe Skills Collide: Measuring Compositional Risk in Agent Skill Ecosystemscited by 14
  2. 2026Capability Safety as Datalog: A Foundational Equivalencecited by 2
  3. 2026Safety as Computation: Certified Answer Reuse via Capability Closure in Task-Oriented Dialoguecited by 0
  4. 2026Silence Is Endorsement: Verification-Status Laundering in LLM Agent Pipelinescited by 0