The second publication for the TIDE project. It has received the Best Paper Award at NOMS 2018.

Snowshoe spam is a type of spam which is notoriously hard to detect. Differently from regular spam, snowshoe spammers distribute the volume among many hosts, in order to make detection harder. To be successful, however spammers need to appear as legitimate as possible, for example, by adopting email best practice like Sender Policy Framework (SPF). This requires spammers to register and configure legitimate DNS domains. Previous studies uses DNS data to detect spam. However, this often happens based on passive DNS data. In this paper we take a different approach. We make use of active DNS measurements, covering more than 60% of the namespace, in combination with machine learning to identify malicious domains crafted for snowshoe spam. Our results show that we are able to detect snowshoe spam domains with a precision of more than 93%. Also, we are able to detect a subset of the malicious domain 2?104 days earlier than the spam reputation systems (blacklists) currently in use, which suggest our method can give us a time advantage in the fight against spam. In a real-life scenario, we have shown that our results allow spam filter operators to block spam that would otherwise bypass their mail filter. A Realtime Blackhole List (RBL) based on our approach is currently deployed in the operational network of a major Dutch ISP.

DDoS attacks threaten Internet security and stability, with attacks reaching the Tbps range. A popular approach involves DNS-based reflection and amplification, a type of attack in which a domain name, known to return a large answer, is queried using spoofed requests. Do the chosen names offer the largest amplification, however, or have we yet to see the full amplification potential? And while operational countermeasures are proposed, chiefly limiting responses to ‘ANY’ queries, up to what point will these countermeasures be effective? In this paper we make three main contributions. First, we propose and validate a scalable method to estimate the amplification potential of a domain name, based on the expected ANY response size. Second, we create estimates for hundreds of millions of domain names and rank them by their amplification potential. By comparing the overall ranking to the set of domains observed in actual attacks in honeypot data, we show whether attackers are using the most-potent domains for their attacks, or if we may expect larger attacks in the future. Finally, we evaluate the effectiveness of blocking ANY queries, as proposed by the IETF, to limit DNS-based DDoS attacks, by estimating the decrease in attack volume when switching from ANY to other query types. Our results show that by blocking ANY, the response size of domains observed in attacks can be reduced by 57%, and the size of most-potent domains decreases by 69%. However, we also show that dropping ANY is not an absolute solution to DNS-based DDoS, as a small but potent portion of domains remain leading to an expected response size of over 2,048 bytes to queries other than ANY.

The DNS TXT resource record is the one that without doubt provide users with the most flexibility of content, as it is a largely unstructured. Although it might be the ideal basis for storing any form of text-based information, it also poses a security threat, as TXT records can also be used for malicious and unintended practices. Yet, we reckon that TXT records are often overlooked in security research. In this paper, we present the first structured study of the uses of TXT records, with a specific focus on security implications. We are able to classify over 99.54% of all TXT records in our dataset, finding security issues including accidentally published private keys and exploit delivery attempts. We also report our lessons learned while dealing with a large-scale, systematic analysis of TXT records.

The possibility to include Unicode characters in domain names allows users to deal with domains in their regional languages. This is done by introducing Internationalized Domain Names (IDN). However, the visual similarity between different Unicode characters - called homoglyphs - is a potential security threat, as visually similar domain names are often used in phishing attacks. Timely detection of suspicious homograph domain names is an important step towards preventing sophisticated attacks, since this can prevent unaware users to access those homograph domains that actually carry malicious content. We therefore propose a structured approach to identify suspicious homograph domain names based not on use, but on characteristics of the domain name itself and its associated DNS records. To achieve this, we leverage the OpenINTEL active DNS measurement platform, which performs a daily snapshot of more than 65% of the DNS namespace. In this paper, we first extend the existing Unicode homoglyph tables (confusion tables). This allows us to detect on average 2.97 times homograph domains compared to existing tables. Our proactive detection of suspicious IDN homograph domains provides an early alert that would help both domain owners as well as security researchers in preventing IDN homograph abuse.

The fourth publication for the TIDE project. The FIRST talk (see here) has been extended into a journal paper for Digital Threats: Research and Practice (DTRAP). In this paper we argue that we, as a security community, should move towards proactive security. However, we shed light on both sides of the coin. We think the ‘optimal’ way is to combine the reactive and proactive methods, to make use of the best of both worlds.