AI Agents Vulnerable to Attacks Embedded in Token Metadata, Warns Blockaid

Blockchain security firm Blockaid has warned that AI agents are vulnerable to attacks embedded in token metadata. The company said attackers can embed prompt-injection instructions in token names, symbols, and descriptions.

This risk is greater when an AI agent has access to a crypto wallet or transaction-execution permissions. In a blog post published on September 29, Blockaid cited several incidents involving AI systems, including an AI-controlled wallet that transferred $215,000 in tokens after being prompted publicly through social media.

Blockaid also pointed out that the increasing number of newly created crypto assets adds to the volume of information that automated systems may process. An agent monitoring large numbers of assets could encounter substantial amounts of token metadata without adequate filtering, potentially allowing manipulated text to enter an AI-driven trading workflow.