Expert commentary: How quantum can make AI safer

The fundamental technology in the world of artificial intelligence (AI) is machine learning. This helps machines such as computers learn from data to perform tasks or make predictions.

Machine learning (ML) powers many everyday technologies, from traffic prediction on your phone and streaming recommendations on your TV to chatbots and self-driving cars. However, challenges include security vulnerabilities, data manipulation, and high energy costs.

However, combining ML with the greater processing power of quantum computing has the potential to make the technology more reliable and “robust.”

Even just a few years ago, quantum robustness was just an idea. Since then, the research undertaken by CSIRO and others promises that this very new frontier in computing will make AI safer, faster and more reliable.

All quotes below can be used in the media. These are from Professor Muhammad Usman, CSIRO’s Quantum Systems Team Leader and editor of Quantum Robustness in Artificial Intelligence, the first book to focus on quantum-enhanced reliability in machine learning systems.

Where do we use machine learning on a daily basis?

ML focuses on learning patterns and improving performance over time. Enabling systems to learn from experience allows them to predict outcomes, understand audio and images, and support complex decision-making. These capabilities are increasingly shaping how organizations operate across a variety of sectors, from healthcare and energy to finance, transportation, and national security.

However, there are several challenges to deploying machine learning in autonomous systems at scale. You may be vulnerable to cyber-attacks and manipulation of your training data. Their energy consumption is enormous. And that authenticity lacks trust.

What is quantum machine learning and how does it work?

Let’s take a step back and talk about quantum computing. The computers and smartphones we use every day use what is known as traditional or “classical” computing. At the most basic level, we process information using bits with a value of 0 or 1, much like a light switch that is turned on or off. All calculations, no matter how complex, will eventually be broken down into long sequences of these values.

Quantum computing, on the other hand, works completely differently. Instead of bits, we work with quantum bits (or qubits), which can take advantage of special properties of the quantum world. Unlike classical bits, qubits can be in the states of 0, 1, partially 0 and 1, or any combination in between.

This proprietary quantum property, known as superposition, allows quantum computers to store and process extremely large datasets at unprecedented speeds. Imagine something like a coin spinning in the air instead of lying flat like heads or tails. While spinning, it has both sides, not just one or the other. Similarly, qubits can explore many possibilities at once before deciding on an answer.

Quantum machine learning is the design and development of new ML models that explicitly exploit the special properties of quantum computing.

What happens when machine learning is integrated with the superior processing power of quantum computing?

With quantum ML, new algorithms are designed and trained using the unique properties of quantum science and the added power of quantum computing. This is a big paradigm shift.

CSIRO’s goal is to design new quantum ML systems that can provide faster training, lower energy consumption, higher accuracy, robust decision-making, and resistance to cyber-attacks.

These quantum ML models promise to create solutions for humanity that are not possible with today’s classical computers.

Why is machine learning safety an issue?

Unfortunately, machine learning is vulnerable to noise, data poisoning, and spoofing attacks. Even highly efficient and well-trained models can be easily fooled by subtle and subtle changes to the dataset. This is extremely dangerous for many applications, such as defense and national security, healthcare, and autonomous vehicles, where robustness is paramount.

For example, self-driving cars can misinterpret red lights as green lights, potentially causing serious accidents and road safety issues. In healthcare systems, weak machine learning models can miss diagnosis of deadly diseases, resulting in significant harm to patients.

Existing classical solutions rely on improving the training of machine learning models to make them robust, but these methods are expensive and often insufficient to fully resolve vulnerabilities. Without better solutions, science needs new approaches to make AI safer. That’s where quantum ML comes in.

What does the research say about quantum machine learning and safer AI?

Quantum ML is poised to become a game-changing technology. This provides a completely different way of processing datasets and learning features to make decisions.

Let’s look at an image as an example. Traditional machine learning processes image data at the pixel level. Thanks to its quantum properties, quantum machine learning processes images at the feature level, i.e. many pixels at a time. Therefore, small, deliberate manipulations of data that might fool a normal AI system cannot fool this quantum-based system.

CSIRO has already shown that quantum ML models are highly robust against a variety of adversarial attacks that easily fool traditional models. We are currently building a complete pipeline of quantum ML models that can be deployed in future autonomous systems to ensure reliable decision-making.

What are the future possibilities for quantum machine learning?

Quantum machine learning is a rapidly evolving field and is very likely to be the first use case for quantum computers that has real-world impact.

Quantum machine learning, like other quantum applications, still faces significant challenges before it can be used in the real world. Current quantum computers are small, noisy, resource-intensive for data and error correction, and difficult to train. As a result, most research in quantum ML remains theoretically or simulation-based, with only a few experimental demonstrations to date.

Nevertheless, given the significant advances in both hardware and software, and the ambitious quantum computing roadmap being pursued by many developers around the world, there is strong reason to be optimistic about quantum ML moving from laboratory research to practical workflows in the near future.

What is CSIRO’s role in this area?

CSIRO hosts large-scale national research and development programs focused on computing, sensing, communications and energy. Their work includes advancing quantum ML models and their applications to real-world problems.

In recent years, CSIRO researchers have advanced the field of quantum ML in several areas, including demonstrating superior robustness, optimizing data encoding in quantum states, overcoming hardware noise with partial error correction, and reducing the resource requirements of quantum ML models.

CSIRO’s research program has been expanded to explore new applications in medical diagnostics, financial and fraud detection, and transport and logistics optimization.

CSIRO has also brought together leading international researchers for the first book on quantum robustness in machine learning models, to be published in April 2026.

CSIRO research will continue to focus on pushing the boundaries of quantum ML models for practical and scalable real-world applications, with the aim of supporting real-world applications as quantum technologies mature.