An official website of the United States government
SSH (Secure Shell) is widely used for remote access to systems and cloud services. This access comes with the persistent threat of SSH password-guessing brute-force attacks (BFAs) directed at sshd-enabled devices connected to the Internet. In this work, we present a comprehensive study of such attacks on a production facility (CloudLab), offering previously unreported insight. Our study provides a detailed analysis of SSH BFAs occurring on the Internet today through an in-depth analysis of sshd logs collected over a period of four years from over 500 servers. We report several patterns in attacker behavior, present insight on the targets of the attacks, and devise a method for tracking individual attacks over time across sources. Leveraging our insight, we develop a defense mechanism against SSH BFAs that blocks 99.5% of such attacks, significantly outperforming the 66.1% coverage of current state-of-the-art rate-based blocking while also cutting false positives by 83%. We have deployed our defense in production on CloudLab, where it catches four-fifths of SSH BFAs missed by other defense strategies.
more »
« less
Thimblerig: A Game-Theoretic, Adaptive, Risk-limiting Security System for Cloud Systems
A significant portion of organizations and applications host client facing servers on cloud-based systems. As the first line of access into a system’s services, these clientfacing servers have a significant attack surface from network adversaries. Once compromised, these systems may be used to send spam, mine crypto, launch DDoS attacks, or used for other nefarious purposes. We propose an adaptive moving target defense that uses game theory to optimize the security and cost to the cloud system. This system leverages the fault-tolerant capabilities of cloud systems with large numbers of client facing servers and the virtualization of these client facing servers by strategically crashing random systems. As a result, an attacker who has compromised a system loses access to it and incurs the cost of having to re-compromise the system once they notice it has been lost. This approach drastically limits the amount of time that an attacker can utilize compromised systems and raises the overall investment required for that time. We have demonstrated via simulation a 90% reduction in the amount of time that an attacker has control over a compromised system for realistic scenarios based on previous data collection of live systems. This approach is agnostic to the method of compromise, so it is even effective against zero-day attacks.
more »
« less
- Award ID(s):
- 1853953
- PAR ID:
- 10575523
- Publisher / Repository:
- IEEE
- Date Published:
- ISBN:
- 979-8-3503-2793-9
- Page Range / eLocation ID:
- 1 to 6
- Format(s):
- Medium: X
- Location:
- Seoul, Korea, Republic of
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
null (Ed.)As organizations drastically expand their usage of collaborative systems and multi-user applications during this period of mass remote work, it is crucial to understand and manage the risks that such platforms may introduce. Improperly or carelessly deployed and configured systems hide security threats that can impact not only a single organization, but the whole economy. Cloud-based architecture is used in many collaborative systems, such as audio/video conferencing, collaborative document sharing/editing, distance learning and others. Therefore, it is important to understand that safety risk can be triggered by attacks on remote servers and confidential information might be compromised. In this paper, we present an AI powered application that aims to constantly introspect multiple virtual servers in order to detect malicious activities based on their anomalous behavior. Once the suspicious process(es) detected, the application in real-time notifies system administrator about the potential threat. Developed software is able to detect user space based keyloggers, rootkits, process hiding and other intrusion artifacts via agent-less operation, by operating directly from the host machine. Remote memory introspection means no software to install, no notice to malware to evacuate or destroy data. Conducted experiments on more than twenty different types of malicious applications provide evidence of high detection accuracymore » « less
-
null (Ed.)Blockchain technology has recently gained high popularity in data security, primarily to mitigate against data breach and manipulation. Since its inception in 2008, it has been applied in different areas mainly to maintain data integrity and consistency. Blockchain has been tailored to secure data due to its data immutability and distributive technology. Despite the high success rate in data security, the inability to identify compromised insider nodes is one of the significant problems encountered in blockchain architectures. A Blockchain network is made up of nodes that initiate, verify and validate transactions. If compromised, these nodes can manipulate submitted transactions, inject fake transactions, or retrieve unauthorized information that might eventually compromise the stored data's integrity and consistency. This paper proposes a novel method of detecting these compromised blockchain nodes using a server-side authentication process and thwart their activities before getting updated in the blockchain ledger. In evaluating the proposed system, we perform four common insider attacks, which fall under the following three categories: (1)Those attacks targeting the Blockchain to bring it down. (2) the attacks that attempt to inject fake data into the database. (3) The attacks that attempt to hijack or retrieve unauthorized data. We described how we implement the attacks and how our architecture detects them before they impact the network. Finally, we displayed the attack detection time for each attack and compared our approach with other existing methods.more » « less
-
The remarkable performance of large language models (LLMs) in generation tasks has enabled practitioners to leverage publicly available models to power custom applications, such as chatbots and virtual assistants. However, the data used to train or fine-tune these LLMs is often undisclosed, allowing an attacker to compromise the data and inject backdoors into the models. In this paper, we develop a novel inference time defense, named CLEANGEN, to mitigate backdoor attacks for generation tasks in LLMs. CLEANGEN is a lightweight and effective decoding strategy that is compatible with the state-of-the-art (SOTA) LLMs. Our insight behind CLEANGEN is that compared to other LLMs, back doored LLMs assign significantly higher probabilities to tokens representing the attacker-desired contents. These discrepancies in token probabilities enable CLEANGEN to identify suspicious tokens favored by the attacker and replace them with tokens generated by another LLM that is not compromised by the same attacker, thereby avoiding generation of attacker-desired content. We evaluate CLEANGEN against five SOTA backdoor attacks. Our results show that CLEANGEN achieves lower attack success rates (ASR) compared to five SOTA baseline defenses for all five backdoor attacks. Moreover, LLMs deploying CLEANGEN maintain helpfulness in their responses when serving benign user queries with minimal added computational overhead.more » « less
-
null (Ed.)Fast networks and the desire for high resource utilization in data centers and the cloud have driven disaggregation. Application compute is separated from storage, but this leads to high overheads when data must move over the network for simple operations on it. Alternatively, systems could allow applications to run application logic within storage via user-defined functions. Unfortunately, this ties provisioning and utilization of storage and compute resources together again. We present a new approach to executing storage-level functions in an in-memory key-value store that avoids this problem by dynamically deciding where to execute functions over data. Users write storage functions that are logically decoupled from storage, but storage servers choose where to run invocations of these functions physically. By using a server-internal cost model and observing function execution, servers choose to directly run inexpensive functions, while preferring to execute functions with high CPU-cost at client machines. We show that with this approach storage servers can reduce network request processing costs, avoid server compute bottlenecks, and improve aggregate storage system throughput. We realize our approach on an in-memory key-value store that executes 3.2 million strict serializable user-defined storage functions per second with 100 us response times. When running a mix of logic from different applications, it provides throughput better than running that logic purely at storage servers (85% more) or purely at clients (10% more). For our workloads, it also reduces latency (up to 2x) and transactional aborts (up to 33%) over pure client-side execution.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- https://doi.org/10.1109/NOMS59830.2024.10575857
