The recent research on Internet Protocol-over-Wavelength Division Multiplexing (IP-over-WDM), driven by the explosion of the Internet data traffic, has uncovered many interesting issues, and much progress has been made as well. While transparent WDM optical networks become more and more popular as the basis of Next Generation Internet (NGI) infrastructure, such networks also raise many unique security issues. The existing protection schemes which only consider inadvertent failures and only rely on postmortem detection and reaction are not suitable for providing security assurance for such infrastructures which require timely protection from malicious sabotage as well as inadvertent faults. In addition, as may have been observed from the past practices, providing a particular solution dedicated to each incessantly discovered new problem would not dramatically improve the situation. In this paper, we discuss the diverse efforts reported in the literature to address the security problem in optical networks. Then, we propose a new approach to protect such infrastructure networks, which is to imbed intelligence into the network, such that they can continuously learn from different faults and attacks and selforganize to protect themselves from potential failures caused by malicious attacks or regular reliability problems.