Safety & Security

BunnyOS is purpose-built for secure, autonomous on-chain execution. It is designed to define strict boundaries around what an agent can do, when it can act, and how its behavior is monitored and enforced.

Framework-Level (runtime) BunnyOS enforces security at the execution and reasoning layer—where decisions are made and actions are generated. Agents are designed with minimal default privileges and operate within clearly defined behavioral boundaries.

• Tool-based Permissioning Only explicitly enabled tools are available to each agent. Others remain completely inaccessible. Tools can be hot-swapped per each user request.

• Tool-specific Constraints Enforce slippage limits, token pair allowlists, liquidity minimums, or protocol-specific ceilings.

• Natural Language Guardrails (Tripwire) Users can define high-level behavioral rules in plain language (e.g. “Never spend more than 20% of the portfolio on a single asset”, or “Avoid NFTs or meme coins”). A secondary AI model monitors the agent’s reasoning and intervenes if these constraints are violated.

• Pre-Execution Simulation Before a transaction is signed, the agent simulates the action to preview on-chain effects and confirm it aligns with its task scope and risk parameters.

Protocol-Level At the network layer, the BunnyOS protocol enforces decentralized runtime policies using smart contracts and validator nodes.

• Security Layer & Policeman Nodes All messages between oracles and agents are intercepted and validated by a decentralized set of Policeman Nodes.

• Identity & Policy Enforcement Each agent publishes its operational boundaries (e.g. what types of data it can consume), which are enforced trustlessly at runtime.