Identity and behavior-based secure personalized message classification system

网络钓鱼目前仍然是通过电子消息(即电子邮件、短消息、推文)进行的最具影响力的攻击形式之一。定向带头网络钓鱼带来了更大的挑战，因为它依赖于从在线数据转储(例如，在黑暗网络中)或通过劫持在线账户收集的受害者的亲密知识。最近高调的黑客事件表明了传统反网络钓鱼解决方案的局限性，如垃圾邮件过滤器和反病毒系统。这些方法主要侧重于分析“信息”，而不是“信使”。因此，很容易通过制作符合预期消息传递标准的“消息”来规避它们(例如，通过删除垃圾邮件的特性和关键字、校对消息以及使用熟悉的术语和名称)。我们开发了一种新的方法来防御定向矛头网络钓鱼攻击，其基础是验证信使身份的真实性，并检查关键消息特征是否与声称的信使身份相对于收件人的消息传递行为匹配。信使身份的真实性是通过文体作者身份验证来确定的，众所周知，这在处理短的非结构化文本时非常具有挑战性。消息传递行为基于消息来源位置模式(即该信使发送真实消息的典型位置)、文件附件模式(即信使附加某些特征的文件的可能性)和嵌入URL模式(即信使嵌入URL的可能性)。提出的模型用于验证传入和传出消息。通过检查传出消息，可以确定合法帐户是否已被劫持，并被利用来向帐户所有者已知的受害者发送网络钓鱼消息。检查传入邮件可确定发件人是否真实且收件人是否知道，从而防止网络钓鱼攻击。拟议项目的目标是进一步开发和广泛测试完全工作的原型，以确定该技术的商业价值，并预期将其转让给现有的网络安全公司。

Phishing is still currently one of the most impactful forms of attack delivered through electronic messages (i.e. emails, short messages, tweets). Targeted spearheaded phishing poses even more challenges as it relies on intimate knowledge of the victim gathered from online data dumps (e.g. in the dark web) or by hijacking online accounts. Recent high-profile hacking incidents have shown the limits of traditional anti-phishing solutions, such as spam filters and anti-virus systems. Those methods are heavily focused on analyzing the "message" and not the "messenger". As such it is easy to evade them by crafting the "message" to fit expected messaging standards (e.g. by removing spam idiosyncrasies and keywords, proof-reading the message, and using familiar terms and names). We have developed a new approach to protect against targeted spearheaded phishing attacks based on verifying the genuineness of the messenger's identity, and checking whether key message characteristics match the messaging behavior of the claimed messenger identity with respect to the recipient. The genuineness of the messenger's identity is established through stylometric authorship verification, which is known to be very challenging when dealing with short unstructured text. Messaging behavior is based on message origin location patterns (i.e. typical locations from where genuine messages were sent by this messenger), file attachment patterns (i.e. likelihood for the messenger to attach files of certain characteristics), and embedded URLs patterns (i.e. likelihood for the messenger to embed URLs). The proposed model is used to verify both incoming and outgoing messages. Checking outgoing messages allows determining whether a legitimate account has been hijacked and is being leveraged to send out phishing messages to victims known by the account owner. Checking incoming messages protects against phishing attacks by establishing whether the sender is genuine and known by the recipient. The goal of the proposed project is to further develop and extensively test a fully working prototype, in order to establish the commercial value of the technology in anticipation of transferring it to existing cybersecurity companies.**********