PraisonAI - Information Disclosure via Shared MultiAgentLedger State

An attacker registers an agent using an ID that duplicates an existing agent's ID. Because the MultiAgentLedger does not validate uniqueness, the new agent is assigned the same ledger instance as the original agent. The attacker can then read the original agent's system prompts and conversation history through the shared ledger, leading to information disclosure [ref_id=1]. No authentication or network-level bypass is required beyond the ability to register an agent with a chosen ID.

The vulnerability resides in the **MultiAgentLedger** component of PraisonAI before version 1.5.115. The component fails to enforce uniqueness of agent IDs, allowing multiple agents to register with duplicate IDs and consequently share the same ledger instance. This shared ledger exposes system prompts and conversation history between agents that should remain isolated.

The patch enforces agent ID uniqueness within the MultiAgentLedger, ensuring that each agent ID maps to exactly one ledger instance. By rejecting registration attempts with duplicate IDs, the fix prevents agents from sharing ledger data and thereby blocks unauthorized access to system prompts and conversation history. The advisory does not include a specific diff, but the remediation guidance confirms that uniqueness validation was added.