Role-Based Deception in Enterprise Networks

Iffat Anjum, Mu Zhu, Isaac Polinsky, William Enck, Michael K. Reiter, Munindar P. Singh

TL;DR

This paper introduces HoneyRoles, a novel deception technique for enterprise networks that uses "honey connections" to mislead adversaries who have compromised network devices. By creating fake network traffic associated with specific organizational roles, HoneyRoles aims to make it harder for attackers to identify high-value targets and dissuade them from acting on potentially false intelligence, thereby improving network security.

What Problem Does This Paper Address?

Modern enterprise networks are increasingly vulnerable to sophisticated attacks, often starting with adversaries gaining a foothold by compromising network devices like routers and switches. From these vantage points, attackers can passively observe network traffic to identify high-value targets such as IT administrators or finance personnel. Traditional defenses are often tuned to detect active reconnaissance, but passive observation is harder to detect. Furthermore, as network devices become more complex, especially with the rise of Software-Defined Networking (SDN), they present more opportunities for compromise and sophisticated attacks like injecting malicious content or blackholing traffic. This paper addresses the challenge of defending against adversaries who have already compromised network forwarding devices and are using passive reconnaissance to plan their next moves.

Methods

The core of HoneyRoles is its use of deception to confuse and deter attackers. The system operates on several key principles:

  • Role-Based Deception: HoneyRoles categorizes client hosts based on their organizational roles (e.g., IT administrator, finance personnel). It then creates "honey connections" – fake network traffic – that mimic the behavior of real users in these roles. This creates metaphorical "haystacks" around the actual network activities of high-value individuals.
  • Honey Connections as Canaries: These fake connections are designed to act as network "canaries." If an adversary interacts with or manipulates a honey connection, HoneyRoles can detect this activity, signaling a potential compromise.
  • Dynamic Forwarding Paths: Leveraging SDN capabilities, HoneyRoles dynamically adjusts the network paths for both real and honey traffic. This makes it harder for an adversary to consistently track specific flows and helps in statistically identifying compromised switches.
  • Belief Maintenance System (BMS): A system component that collects alerts from honey agents and uses statistical inference to rank network switches based on their likelihood of being compromised. This helps administrators focus their remediation efforts.
  • Deception and Dissuasion: The primary goals are to deceive adversaries by providing misleading information and to dissuade them from acting on that information by making the risk of detection high.

Datasets

Not discussed in the paper. The authors used a simulated Fat-Tree network topology for their experiments.

Key Insights and Ideas

  • Shifting the Defense Paradigm: Instead of solely focusing on detection, HoneyRoles introduces deception as a proactive defense mechanism against compromised network devices.
  • Leveraging Organizational Roles: The insight that adversaries target specific roles (like IT admins) is crucial. By mimicking these roles, HoneyRoles can create highly convincing decoys.
  • Honey Connections as Double Agents: These fake connections not only mislead attackers but also serve as early warning systems (canaries) for detecting malicious activity.
  • SDN as an Enabler: The programmability and dynamic control offered by SDN are essential for implementing the dynamic forwarding paths and managing honey connections effectively.

Key Results

  • Detection Accuracy: In simulations using the PRISM probabilistic model checker, HoneyRoles consistently ranked compromised switches among the most suspicious, even when multiple switches were compromised. For a single compromised switch, it was typically ranked in the top 1 or 2.
  • Performance Overhead: Experiments using Mininet showed that HoneyRoles has a small impact on network request completion time. For 90% of real requests, the overhead was less than 14% compared to a baseline without deception.
  • Robustness: The system demonstrated effectiveness across different switch locations (edge, aggregate, core) and with varying numbers of compromised switches.

Limitations and Future Work

  • Ambient Traffic Generation: The current implementation does not include a sophisticated ambient network traffic generator, which is noted as an area for future work. Such generators are crucial for making honey traffic indistinguishable from real traffic.
  • Attack Variety: While the system demonstrates detection for SSL-stripping and blackholing, adapting to a wider range of sophisticated attacks would require further development of detection algorithms within the honey agents.
  • TCB Assumptions: The system relies on a trusted computing base (TCB) for the SDN controller and southbound network, and assumes host agents are trustworthy.
  • Network Topology: The security analysis used a Fat-Tree topology, which is a common but specific network structure. Evaluating HoneyRoles on more diverse enterprise network topologies would be beneficial.
  • Adversary Awareness: The current model assumes an adversary that is aware of the existence of a deception system but not necessarily its specifics. Future work could explore more advanced adversary models.

Summary updated

BibTeX
@inproceedings{anjum2021role,
  author = {Iffat Anjum and Mu Zhu and Isaac Polinsky and William Enck and Michael K. Reiter and Munindar P. Singh},
  title = {{Role-Based Deception in Enterprise Networks}},
  booktitle = {CODASPY},
  year = {2021},
  doi = {10.1145/3422337.3447824},
}
Figure