An agent is an autonomous software entity capable of proactivity and cognition to reach their goals. A group of agents cooperating or competing between themselves to reach their goals, common or not, is a Multi-agent System (MAS). MAS can also be Open over the Internet, and agents can move from one MAS to another. However, this free flow of agents between MAS can pose a significant security risk to the MAS, as an agent can have malicious intent. This work then presents an approach to provide access control of Open MAS. This approach, inspired by networking firewalls, consists of defining a set of guidances and policies to manage the connection between Open MAS, which can be either to drop or accept messages. If the set of guidances and policies accept a message, it is sent to the agent’s mind to continue its reasoning. Otherwise, it is dropped as if it were never received. We then conduct a proof of concept where 2 MAS - an authorized and a malicious one - try to migrate to a third MAS using bio-inspired migration protocol. This experiment shows that our approach can effectively block the malicious MAS while allowing the authorized MAS to continue.