AgentGuard – Transaction guardrails for AI agents that hold wallets (github.com)

AgentGuard has introduced a new transaction guardrail layer designed for AI agents managing crypto wallets. This system requires agents to submit proposed transactions to AgentGuard, which simulates the transaction against the current blockchain state and evaluates it based on operator-defined policies. The outcome can be to allow, block, or escalate the transaction, accompanied by a clear explanation, a risk score, and an immutable audit trail. This is particularly significant for the AI/ML community as it enhances security in automated transactions by ensuring that AI agents cannot independently determine the safety of transactions. The guardrail operates on Ethereum Virtual Machine (EVM) chains and Solana, utilizing a conservative default policy, a TypeScript SDK, and a comprehensive rule engine that includes a contract reputation layer and an optional behavioral anomaly detection layer. The latter scores transactions based on the agent's historical behavior, thereby identifying potentially fraudulent activities that static rules might miss. As an open-source solution under the Apache License 2.0, it allows developers to integrate these protective measures into their applications, fostering a more secure environment in decentralized finance (DeFi) and beyond. The proactive model of risk assessment and transaction evaluation serves as a crucial advancement in safeguarding automated interactions in the ever-evolving landscape of blockchain technology.