Ad hoc networks are becoming an important research aspect due to the self-organization network, dynamically changing topology, temporary network life and equal relationship among member of nodes. However, all the characters of ad hoc network make the security problem more serious. Network security and trustworthiness become the key problems of the network. Denial-of-service and Black hole attacks are the two puzzles in the security of ad hoc network. There are not satisfied solutions to solve the problem. A novel multi-agent-based dynamic lifetime intrusion detection and a response scheme are proposed to combat the two types of attacks. Multi-agents are related to one route request (RREQ)–route reply (RREP) stream. One agent monitors the nodes in three-hop zone. Agent can periodically update itself by the trustworthiness of the neighbor nodes. It can efficiently improve trustworthiness, decrease computing complexity and save energy consumption for network securities. Agent security specifications have been extracted from the feature of the attacks. Multi-agents can trace RREQ and RREP messages, stream to aggregate the key information to link list and MAC-IP control table and analyze them by intrusion detection algorithm. Different security metrics are proposed to quantitatively evaluate network security performance under different attacks. Ns2 simulator is expanded to validate the security scheme. Simulation results show that a multi-agent-based dynamic lifetime security scheme is highly effective to detect and block the two kinds of attacks.