Tech Insights

‘Brain’ Power: Does AI Increase Smart Grid Cyber Risk?—Part 2

May 28, 2024 by Liam Critchley

How can engineers protect an AI-enabled grid from cyberattacks? This is the second in a three-part series examining the benefits and drawbacks of an AI-enabled smart grid.

Artificial intelligence (AI) can improve smart grid operations, but behind the extra data and analytical capabilities lie a range of potential entry points that are not as robustly protected as established software systems. Using AI presents numerous risks to smart grid operations, but when properly employed, AI can reduce potential power grid cyberattacks in the future.

Preventing cyberattacks begins with understanding smart grid vulnerabilities and learning how AI can detect and stop outside agents from destabilizing the grid, stealing data, and introducing malware.

 

AI can help prevent cyberattacks, but it also makes the grid more vulnerable.

AI can help prevent cyberattacks, but it also makes the grid more vulnerable. Image used courtesy of Adobe Stock

 

How AI Can Deter Cyberattacks 

As smart grids expand and connect the physical world to the digital world, hackers are finding new ways to impact national security and lives, as many businesses and healthcare establishments rely on a consistent, stable electricity flow. 

Hackers can attack smart grids in numerous ways to disrupt or manipulate processes across the whole network, from energy generation to distribution and operations. The interconnected network presents many entry points, making it susceptible to cyberattacks, including confidentiality, integrity, and denial of service attacks.

AI offers ways to identify, detect, and respond to known and unknown threats across energy system networks. First, AI enables machine-speed analysis of information technology and operational technology data to detect, identify, and minimize potential intrusions within energy infrastructure before they cause damage. Second, AI can detect anomalies within networks and understand network packet properties to identify any attacks.

 

Video used courtesy of U.S. Government Connect

 

Cybersecurity Algorithms

While many AI algorithms are being developed for energy cybersecurity applications, machine learning, deep learning, and federated learning (a subset of machine learning) are the frontrunners. AI can also be used to predict attacks in the physical and cyber layers of the interconnected smart grid network. For example, algorithms trained on voltage and current datasets can detect any malicious measurements, and cyber-physical systems can detect time-delay attacks by processing the raw data streams from the smart sensors.

Machine learning algorithms can identify cyberattack stages following the cyber-kill chain by examining the attack signatures, network packets, and phasor measurements. It can also detect anomalous behavior suggesting a system intrusion or malicious activity.

Federated learning is used in a more decentralized capacity to detect false data injection attacks on key renewable energy components, such as solar converters. It can also uncover potential energy theft.

Deep learning algorithms can detect attacks on power transmission lines, such as in protective relays, and recognize time delay attacks against automatic generation control systems. They can find attacks targeting the smart grid communication networks, which enable the many grid components to communicate with each other and the wider grid. Deep learning can expose attacks in Modbus/TCP and DNP3 communication protocols.

Despite AI’s benefits, many detection approaches are still at an academic level and have not matured enough for wider grid use. AI could play a key role in smart grid defense in the future and will work seamlessly with the Internet of Things solutions now in place. 

 

What Cybersecurity Risks Are AI Systems Up Against?

 Since wider AI networks are relatively new, they are a target because they are not as robust against manipulative attacks, known as adversarial attacks, that could turn AI against the power grid.

Additionally, malicious parties could manipulate the grid’s AI to damage smart grid networks and the wider energy infrastructure. 

AI systems are susceptible to traditional vulnerabilities and a new class of AI-specific vulnerabilities. 

 

Smart grid vulnerable areas.

Smart grid vulnerable areas. Image used courtesy of the authors

 

Adversarial Attacks on the Smart Grid

Adversarial attacks can manipulate the algorithms intended to provide positive results and turn them into negative outcomes.

One adversarial attack, the poisoning attack, aims to modify, add, or alter the data used to train the AI algorithm. This causes the AI model to learn the wrong behavior. For example, poisoning could induce misconceptions about what normal system operations should look like and modify the data energy systems use. Altering the data can also create backdoor scenarios where the model doesn’t detect changes in the grid when it should, such as the need for equipment maintenance.

Another potential attack on AI grid networks is evasion attacks. These attacks use input data that is not distinguishable from regular data to the human eye, so they are also based on human error. Adding false data to models changes the model’s output and provides unfavorable outcomes to the operator. One example is manipulating energy market prices to underestimate or overestimate energy prices.

The final adversarial attack is a data extraction attack. These attacks learn sensitive information about the smart grid’s AI model and the training data. This knowledge could enable malicious parties to access highly specific information about the energy system.

Defending against these attack types is a standalone branch of cybersecurity, so it can’t lean on existing strategy frameworks to overcome any potential attack. Additionally, current AI cybersecurity defense techniques are not mature enough to combat such sophisticated attacks. However, in the meantime, best practices surrounding access controls, vigilant human supervision, and data curation methods can all help mitigate any potential risks.

 

AI-on-AI Cyberattacks

The other main cyber risk comes from hackers and malicious parties using AI algorithms to target the electric smart grid’s AI network. AI used in this way can either directly coordinate the attack or plan attacks (both cyber and physical) on energy infrastructure. AI may also make attacks easier to perform with limited data by using its inference capabilities. 

AI can sometimes make cyberattacks easier by enabling a wider range of people (including less technically proficient ones) to carry out attacks. However, in other cases, AI attacks will still require some sophistication. These attacks could be much more destructive and effective than were previously possible without AI assistance.

When used nefariously, AI can help increase the success rate of cyberattacks. AI can search for network vulnerabilities to reduce the time and workforce required to perform an attack. It can also generate a model-based attack design, using information about a specific energy system to design a more complex attack model optimized for the particular infrastructure. This means the attack will have a higher impact.

Attackers can also use AI to gain autonomous control grid devices to perform physical attacks. For example, they could combine AI with unmanned drone systems to perform remote physical attacks on key infrastructure. 

AI-driven malware also presents a defense challenge because it may be able to adapt to the system it is attacking to seek out high-value targets and update its objectives during an attack. Finally, AI could be used to evade detection when performing a cyberattack, enabling attackers to bypass firewalls and shroud themselves from detection.

 

AI’s Future in the Smart Grid

As an ever-developing area, using AI within smart grids has many benefits and risks. While AI as a preventative measure against many types of cyberattacks has potential, the novelty opens the door for hackers to attack. More developments will be coming, and when AI systems are more robust to remove some vulnerabilities, we will likely see them being deployed across many smart grids as a preventative and protective measure.

This is the second of a three-part series on using AI in the smart grid. The third and final installment will evaluate how AI can help integrate renewable energy into the grid.