Artificial intelligence is changing cybersecurity in two opposite directions at the same time. It is giving attackers faster and more convincing ways to deceive people, automate attacks and identify weaknesses. At the same time, it is giving defenders new capabilities to detect threats, investigate incidents and respond at a speed that human teams alone cannot match.
Imagine an employee in the finance department receiving an urgent message from the CFO late at night. The email looks legitimate. A follow-up call comes from a familiar voice. If video is involved, the face may also look completely real. The request relates to something the company is actually working on. Nothing immediately looks like a traditional cyberattack.
This is where AI changes the nature of the threat. Cybersecurity has traditionally focused on protecting systems, accounts and networks. Increasingly, attackers are targeting something more fundamental: human trust.
Phishing is a good example. In the past, many phishing messages were easy to recognize because they were generic, badly written or unrelated to the recipient. Generative AI can produce highly personalized messages in seconds. An attacker can use publicly available information about a company, employee or executive to create communication that appears relevant to a real project, event or professional relationship.
“The important change is not simply better grammar. It is better context.”
AI allows attackers to operate with greater scale, speed and personalization. Activities that once required significant research and manual effort can increasingly be automated. One attacker can potentially create thousands of different messages, each adapted to a different individual.
Deepfakes push this problem further. We are entering an environment where seeing someone is no longer enough to confirm identity, and hearing someone’s voice is no longer sufficient evidence that they are actually speaking.
For organizations, this creates a serious problem because many business processes still depend heavily on recognition. Employees trust familiar voices, familiar names, familiar writing styles and familiar faces. AI can increasingly imitate all of them.
“But AI is not only strengthening the attacker.”
Security teams are also using AI to process enormous volumes of information that would be impossible for humans to examine manually. A modern security operations centre may receive thousands of alerts every day. AI can help analysts identify unusual behaviour, connect events across different systems, summarize incidents and prioritize the situations that require immediate investigation.
Consider an analyst reviewing hundreds of security events. Instead of manually examining every login, process and network connection, an AI assistant may quickly identify that a successful login was followed by a suspicious process, an unusual external connection and repeated authentication failures. The analyst still makes the final judgment, but AI can dramatically reduce the time needed to understand what happened.
This is perhaps the strongest defensive value of AI. It does not necessarily replace cybersecurity professionals. It allows them to spend less time processing information and more time making important decisions.
There is also a third dimension to the problem. Once organizations begin using AI extensively, AI systems themselves become part of the attack surface.
An AI assistant may have access to internal documents, customer information, email, databases or cloud services. Future AI agents may be able not only to recommend actions but also to perform them. They may send messages, update records, generate code, modify files or interact with business systems.
The difference between a chatbot and an AI agent is therefore important. A chatbot gives an answer. An agent can take an action.
When an AI system can act, organizations must think about it almost like another digital employee. What information can it access? What actions can it perform? Who approves those actions? What happens if the system is manipulated? Can its access be immediately removed?
This also introduces risks such as prompt injection, where malicious instructions hidden inside content can influence the behaviour of an AI system. An employee may receive a malicious document without realizing that the intended target is not the employee at all, but the AI assistant that will eventually read or process the document.
Cybersecurity therefore has to expand again. Protecting networks and passwords is no longer enough. Organizations must also protect AI models, the information they consume, the tools they can access and the actions they are allowed to perform.
For Canadian technology leaders, the response should not be to avoid AI. Attackers will continue adopting these technologies regardless of whether responsible organizations do the same.
Instead, companies need to redesign security around verification.
A familiar voice should not authorize a financial transfer. A realistic video call should not automatically confirm identity. A confident AI-generated recommendation should not automatically become a business decision. High-risk actions need independent verification, clear authorization and appropriate human oversight.
Organizations also need to rethink cybersecurity awareness training. Employees have traditionally been taught to identify obvious warning signs such as unusual spelling, suspicious links or strange email addresses. Those indicators are becoming less reliable. Training now needs to focus more strongly on verifying unexpected requests, especially those involving money, credentials, sensitive information or urgent changes to normal procedures.
The same principle applies to AI systems. If an AI agent can perform important actions, it should receive only the minimum access necessary. Its activity should be monitored, logged and limited. Critical actions should require additional approval.
The cybersecurity competition is therefore changing. We are moving toward a world where AI-assisted attackers increasingly face AI-assisted defenders. One system may search for vulnerabilities while another searches for the same weaknesses before they can be exploited. One AI may generate a highly convincing fraudulent message while another analyzes that message for signs of manipulation.
“Yet the central issue is not ultimately artificial intelligence. It is trust.”
Return to the finance employee receiving the urgent request from the CFO. The company’s firewall may have worked perfectly. The password may never have been stolen. The computer itself may not have been compromised. But the employee believed the wrong person. Trust was compromised.
That is why the cybersecurity challenge of the AI age is no longer simply keeping machines secure. It is deciding what and whom we can trust when machines can generate, imitate, reason, decide and increasingly act.
“When intelligence becomes artificial, trust must become engineered.”
Hamed Taherdoost, PhD, is Associate Vice President, Academic and Full Professor at University Canada West. An award-winning researcher and R&D leader with over 25 years of international academic and industry experience, his work spans AI, cybersecurity, innovation, and research and development.
Leave a Reply