Elif Ustundag Soykan

We propose a novel intent-based method to prevent security attacks on the safety functions in cyber-physical systems used in smart manufacturing. Context information about a cyber-physical system is collected from various sensors including non-safety sensors measuring device temperature, motor rotation speed, or instantaneous power consumption of machines. Such contextual information along with operational and business intent of the system under consideration are then used to check whether the current situation is indeed an emergency situation or a normal situation. Unlike the conventional safety systems that only rely on raw sensor data and safety protocol status packets from safety sensors, which might be spoofed and/or modified, decision on the safety situation in our method is intelligently made by comparing aggregated sensor information from the cyber-physical system and its environment for compliance with pre-configured operational intents that define the normal safe and secure operation of the system. We also show how to integrate Machine Learning (ML) and Artificial Intelligence (AI) into the proposed method for efficient and automated analysis of both intents and aggregated context information to make more intelligent decisions in execution of functional safety protocols. Our proposed AI/ML integration approach also enables the prediction of safety critical situations before they occur. The proposed method aims to prevent unnecessary switching to fail-safe mode causing insecure system states such as emergency doors opening or system halting in normal situations. In addition, it prevents sticking in non-safe states and not switching to fail-safe mode in real emergencies which could cause hazard on device and/or people.

The increasing transformation from the legacy power grid to the smart grid brings new opportunities and challenges to the power system operations. Bidirectional communications between home-area devices and the distribution system empower smart grid functionalities. More granular energy consumption data flows through the grid and enables better smart grid applications. This may also lead to privacy violations since the data can be used to infer the consumer’s residential behavior, so-called power signature. Energy utilities mostly aggregate the data especially if the data is shared with stakeholders for the management of market operations. Although this is a privacy-friendly approach, recent works show that this does not fully protect privacy. On the other hand, some applications like nonintrusive load monitoring require disaggregated data. Hence, the challenging problem is to find an efficient way to facilitate smart grid operations without sacrificing privacy. In this paper, we propose a privacy-preserving scheme that leverages consumer privacy without reducing accuracy for smart grid applications like load monitoring. In the proposed scheme, we use a trusted execution environment (TEE) to protect the privacy of the data collected from smart appliances (SAs). The scheme allows customer-oriented smart grid applications as the scheme does not use regular aggregation methods but instead uses customer-oriented aggregation to provide privacy. Hence the accuracy loss stemming from disaggregation is prevented. Our scheme protects the transferred consumption data all the way from SAs to Utility so that possible false data injection attacks on the smart meter that aims to deceive the energy request from the grid are also prevented. We conduct security and game-based privacy analysis under the threat model and provide performance analysis of our implementation. Our results demonstrate that the proposed method overperforms other privacy methods in terms of communication and computation cost. The execution time of aggregation for 10,000 customers that each has 20 SAs is approximately 1 second. The decryption operations performed on the TEE have a linear complexity e.g., 172800 operations take around 1 second while 1728000 operations take around 10 seconds. These results can scale up using cloud or hyper-scalers for real-world applications as our scheme performs offline aggregation.

As machine learning and artificial intelligence (ML/AI) are becoming more popular and advanced, there is a wish to turn sensitive data into valuable information via ML/AI techniques revealing only data that is allowed by concerned parties or without revealing any information about the data to third parties. Collaborative ML approaches like federated learning (FL) help tackle these needs and concerns, bringing a way to use sensitive data without disclosing critically sensitive features of that data. In this paper, we provide a detailed analysis of state of the art for collaborative ML approaches from a privacy perspective. A detailed threat model and security and privacy considerations are given for each collaborative method. We deeply analyze Privacy Enhancing Technologies (PETs), covering secure multi-party computation (SMPC), homomorphic encryption (HE), differential privacy (DP), and confidential computing (CC) in the context of collaborative ML. We introduce a guideline on the selection of the privacy preserving technologies for collaborative ML and privacy practitioners. This study constitutes the first survey to provide an in-depth focus on collaborative ML requirements and constraints for privacy solutions while also providing guidelines on the selection of PETs.

The increasing transformation from the legacy power grid to smart grid brings new opportunities and challenges to the power system operations. Advanced metering infrastructure makes bidirectional communications possible between home-area devices and the distribution system which also empowers metering and monitoring functionalities. This means that more information related to consumers flows through the grid. Hence, enabling privacy is an inevitable requirement for advanced metering operations. On the other hand, fulfilling the requirement brings additional communication and computational cost. The challenging problem is to find an efficient way to facilitate advanced metering operations without sacrificing privacy. In this paper, we propose a privacy-preserving scheme that leverages consumer privacy without reducing accuracy for smart grid applications like load profiling. In the proposed scheme, we collect consumption data from smart appliances (SAs) and transfer it to a trusted environment disallowing the smart meter or other intermediary interference. This scheme allows distribution system operators to provide consumers added value services based on their load profile on SA level without violating their privacy. We conduct security and game-based privacy analysis under the threat model, and provide performance analysis of our implementation.

In recent years False Base Stations (FBSs) have received increased attention. A False Base Station can perform active or passive attacks against mobile devices or user equipment (UE) to steal private information, such as International Mobile Subscriber Identifier (IMSI), to trace users locations, or to prevent users from getting service from operators. Most of the existing solutions related to FBS have focused on the detection aspects of the false station rather than locating its position. However, once an FBS is detected in a network, discovering its exact location precisely and remotely becomes highly crucial to initiate preventive actions. In this work, we propose a network-based localization method for estimating the exact geographical position of an FBS whose existence is already detected in a cellular network. Our method relies on a comparative pairwise analysis of the Reference Signals Received Power (RSRP) values reported as a standard procedure by the UEs in the vicinity of FBS through their measurement reports. Specifically, for each pair of related measurement reports, we identify a half-plane indicating the probable location of the FBS and then predict the exact location based on the intersection of all obtained half-planes. We have implemented and experimentally evaluated our proposed method in the Network Simulator 3 ( ns-3 ) and showed that it accurately estimates FBS location with meter-level precision under different scenarios in a cellular network.

Demand response (DR) is a vital element for a reliable and sustainable power grid. Consumer behavior is a key factor in the success of DR programs. In this study, we focus on how consumer reaction to Short Messaging Service (SMS) messages can disturb the demand response. We present a new type of threat to DR programs using SMS phishing attacks. We follow a holistic approach starting from a risk assessment focusing on DR programs’ notification message security following the Smart Grid Information Security (SGIS) risk methodology. We identify threats, conduct impact analysis, and estimate the likelihood of the attacks for various attacker types and motivations. We implemented deterministic and randomized attack scenarios to demonstrate the success of the attack using a state-of-the-art simulator on the IEEE European Low Voltage Feeder Test System. Simulations show that the attack results in local outages, which may lead to large-scale blackouts with the cascading effect on the power system. We conclude that this is a new type of threat that has been overlooked, and it deserves more attention as mobile devices will continually be part of our lives.

As the role of information and communication technologies gradually increases in our lives, software security becomes a major issue to provide protection against malicious attempts and to avoid ending up with noncompensable damages to the system. With the advent of data-driven techniques, there is now a growing interest in how to leverage machine learning (ML) as a software assurance method to build trustworthy software systems. In this study, we examine how to predict software vulnerabilities from source code by employing ML prior to their release. To this end, we develop a source code representation method that enables us to perform intelligent analysis on the Abstract Syntax Tree (AST) form of source code and then investigate whether ML can distinguish vulnerable and nonvulnerable code fragments. To make a comprehensive performance evaluation, we use a public dataset that contains a large amount of function-level real source code parts mined from open-source projects and carefully labeled according to the type of vulnerability if they have any. We show the effectiveness of our proposed method for vulnerability prediction from source code by carrying out exhaustive and realistic experiments under different regimes in comparison with state-of-art methods.