ElizaOS: planted ‘memories’ can steer crypto agents
- Entry
- AGT-005
- When
- May 2025
- Type
- Demonstration
- Evidence
- D Grade D
- Lost
- None (research)
- Who
- Princeton University and Sentient Foundation researchers
- The agent
- ElizaOS, an open-source framework for agents that operate on blockchains
- Its authority
- Agents built on it can trade and transfer
- How it was gamed
- Researchers inserted false memories into an agent’s memory store from another platform. The agent later recalled them and acted on them, which the authors say could enable unauthorized transfers.
- What was lost
- No loss reported; research.
- The design gap
- Stored memories were treated as trustworthy instructions.
- Principles
- Evidence
- Grade D. Research paper, reported by Decrypt.
- Source
Cite this entry: Monetal Labs, Gamed AI Agents Register v0.2, entry AGT-005. https://monetalabs.io/register/agt-005