Giving AI Agents Access to Cryptocurrency and Smart Contracts Creates New Vectors of AI Harm
cs.AI, cs.CR
Submitted: 2025-07-11
Updated: 2026-08-26
Code: https://github.com/circlefin/stablecoin-evm
Project page: https://olas.network/services/prediction-agents
License: http://creativecommons.org/licenses/by/4.0/
The gist: There is growing interest in giving AI agents access to cryptocurrencies and smart contracts.
Terminology
Abstract
There is growing interest in giving AI agents access to cryptocurrencies and smart contracts. But doing so, this position paper argues, could spawn categorically new vectors of AI harm. To support this notion, we first examine the unique technical properties of cryptocurrencies and smart contracts that, when paired with AI agents, make this so. Next, we describe three of these new vectors of AI harm in detail, providing a first-of-its-kind taxonomy. Finally, we call for more research into ways to prevent or mitigate these risks, proposing several as a starting point.
Sources
- Frontier AI Regulation: Managing Emerging Risks to Public Safety
- On the Opportunities and Risks of Foundation Models
- Large Language Models as Tool Makers
- Efficacy of Various Large Language Models in Generating Smart Contracts
- Coinvisor: An RL-Enhanced Chatbot Agent for Interactive Cryptocurrency Investment Analysis
- LLM Agents can Autonomously Hack Websites
- AI Agent Smart Contract Exploit Generation
- CryptoBench: A Dynamic Benchmark for Expert-Level Evaluation of LLM Agents in Cryptocurrency
- AI Agents That Matter
- LLM-Powered Multi-Agent System for Automated Crypto Portfolio Management
- The Alignment Problem from a Deep Learning Perspective
- Feedback Loops With Language Models Drive In-Context Reward Hacking
- Position: Key Claims in LLM Research Have a Long Tail of Footnotes
- Real AI Agents with Fake Memories: Fatal Context Manipulation Attacks on Web3 Agents
- TPTU: Large Language Model-based AI Agents for Task Planning and Tool Usage
- The Feasibility of a Smart Contract "Kill Switch"
- Tools Fail: Detecting Silent Errors in Faulty Tools
- Model evaluation for extreme risks
- Cross Chain Bribery Contracts: Majority vs Mighty Minority
- Eliza: A Web3 friendly AI Agent Operating System
Related papers
- MAVEN-T: Reinforced Heterogeneous Distillation for Real-Time Multi-Agent Trajectory Prediction
- Model Discovery Agent: LLM-assisted Bayesian experiment design for data-efficient discovery of mechanistic world models
- The Clinician's Veto: Navigating Trust, Liability, and Uncertainty in Autonomous AI Prescribing
- MindHelper: Closed-Loop Embodied Mental-State Reasoning for Precision Intervention
- Incumbent Advantage: Brand Bias and Cognitive Manipulation Dynamics in LLM Recommendation Systems
- VSAL: A Vision Solver with Adaptive Layouts for Graph Property Detection