D2M: Dynamic Defense and Modeling of Adversarial Movement in Networks

Scott Freitas, Andrew Wicker, Duen Horng Chau, Joshua NeilarXiv 2020 · arXiv · OpenReview

TL;DR

This paper introduces D2M, a novel framework for quantifying network vulnerability to lateral attacks and identifying at-risk devices. By modeling adversarial movement using real-world attack strategies and a Monte-Carlo approach on authentication graphs, D2M empowers IT administrators to build more robust credential policies and defenses.

What Problem Does This Paper Address?

In today's interconnected enterprise networks, a single compromised credential can unlock a cascade of "lateral movement" attacks, allowing adversaries to escalate privileges and gain control of the entire network. However, studying these attacks is challenging due to their stealthy nature, difficulty in obtaining ground truth, and the fact that attackers already possess valid credentials. Existing methods often detect lateral movement after it occurs or rely on simplified models. This paper tackles the critical need for a proactive approach: quantifying a network's inherent vulnerability to such attacks and identifying specific machines that are prime targets, enabling defenders to prioritize their efforts.

Methods

The D2M framework is built around three core components:

  1. Authentication Graph Construction: The system transforms raw device authentication logs (like Windows NTLM or Kerberos) into a directed, unweighted graph. Each node represents a machine, and an edge signifies an authentication event. Crucially, this graph incorporates domain knowledge about credential caching, network hygiene (levels of privilege distribution), and the role of domain controllers.
  2. Lateral Attack Modeling: D2M models adversarial movement using three distinct attack strategies:
  3. Random Walk-Explore (RWE): A black-box strategy assuming no prior network knowledge, mimicking random exploration.
  4. Rank-Explore (RE) & Degree-Explore (DE): Gray-box strategies that leverage network topology information (PageRank for RE, node degree for DE) to guide the attacker more intelligently. These strategies generate "attack paths" – sequences of compromised machines leading towards the domain controller.

  5. Vulnerability Quantification & Defense Identification:

    • Network Vulnerability: A novel Monte-Carlo simulation is used to estimate network vulnerability as a probabilistic function of the network topology, credential distribution, and initial penetration point. This results in a score between 0 and 1, indicating how susceptible the network is to lateral attacks.
  6. At-Risk Machine Identification: Five graph mining techniques are proposed to identify machines most vulnerable to attack. A novel method, AnomalyShield (AS), is introduced, which prioritizes machines with anomalous neighbors and high eigencentrality, combining static and dynamic defense insights.

Datasets

The framework was evaluated using three real-world authentication graphs:

  • Two anonymized enterprise networks from Microsoft (Gs and Gi).
  • One open-sourced network from Los Alamos National Laboratory (LANL).

These datasets contained up to 223,399 authentications, providing a realistic basis for evaluating the D2M framework.

Key Insights and Ideas

  • Proactive Defense is Crucial: The paper strongly advocates for moving beyond reactive detection to proactively quantifying and mitigating network vulnerability to lateral movement.
  • Domain Knowledge is Key: Integrating real-world adversarial actions and IT security concepts (like credential caching and network hygiene) into graph models significantly enhances their predictive power.
  • Informed Attackers are Faster: Gray-box attack strategies (RE, DE) that leverage network topology information lead to shorter attack paths, making them harder to detect. This highlights the importance of understanding attacker behavior.
  • Network Hygiene Matters: Increasing the "hygiene" of a network (i.e., reducing the prevalence of high-privilege credentials) demonstrably reduces vulnerability and lengthens attack paths, empowering IT admins to tune their policies.
  • AnomalyShield Shows Promise: The novel AnomalyShield defense strategy appears to be a robust general defense, performing well across different network topologies and attacker knowledge levels by combining centrality and anomaly detection.

Key Results

  • D2M successfully models lateral attack paths and quantifies network vulnerability using real-world data.
  • Increasing network hygiene levels (from h1 to h3) significantly reduced vulnerability scores for the tested graphs, with some networks achieving zero vulnerability at the highest hygiene level.
  • Informed attack strategies (RE, DE) consistently resulted in shorter attack paths compared to the black-box RWE strategy.
  • The AnomalyShield defense strategy demonstrated strong performance in identifying at-risk machines across various scenarios, outperforming other tested defense strategies in many cases.

Limitations and Future Work

  • Computational Budget: For the Glant dataset, certain combinations of defense strategies and hygiene levels exceeded the computation budget, limiting the scope of those specific results.
  • Attack Path Modeling: The paper focuses on reaching the domain controller and leaves the modeling of the "exfiltration" phase of the attack kill chain to future work.
  • Partial Information Defense: The defense strategies currently assume the defender has full visibility into anomalous activity. Future work could explore defense strategies under partial information scenarios.
  • Generalizing Vulnerability: While the current vulnerability score focuses on reaching the domain controller, future work could generalize this to other high-value targets within the network.

Summary updated

Abstract

Given a large enterprise network of devices and their authentication history (e.g., device logons), how can we quantify network vulnerability to lateral attack and identify at-risk devices? We systematically address these problems through D2M, the first framework that models lateral attacks on enterprise networks using multiple attack strategies developed with researchers, engineers, and threat hunters in the Microsoft Defender Advanced Threat Protection group. These strategies integrate real-world adversarial actions (e.g., privilege escalation) to generate attack paths: a series of compromised machines. Leveraging these attack paths and a novel Monte-Carlo method, we formulate network vulnerability as a probabilistic function of the network topology, distribution of access credentials and initial penetration point. To identify machines at risk to lateral attack, we propose a suite of five fast graph mining techniques, including a novel technique called AnomalyShield inspired by node immunization research. Using three real-world authentication graphs from Microsoft and Los Alamos National Laboratory (up to 223,399 authentications), we report the first experimental results on network vulnerability to lateral attack, demonstrating D2M's unique potential to empower IT admins to develop robust user access credential policies.
BibTeX
@misc{freitas2020d,
  author = {Scott Freitas and Andrew Wicker and Duen Horng Chau and Joshua Neil},
  title = {{D2M: Dynamic Defense and Modeling of Adversarial Movement in Networks}},
  year = {2020},
  eprint = {2001.11108},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2001.11108},
}
Figure