In the AI era, phishing remains the top attack vector in both use (17% of breaches) and breach costs (averaging at $5.29 per breach) according to the IBM Cost of a Data Breach Report 2026. AI creates more personalized content for phishing with higher success rates. At the same time, AI can identify vulnerabilities, use the phished entry point and elevate privileges for lateral movement at greater speed.
None of this is new – cyberattacks haven’t changed with AI. “It just allows attackers to do the same familiar things much faster and at greater scale than ever before. The pressure comes from minimizing the time for defenders to respond,” says Steve Wallin, director of core integration and R&D labs leader at IBM UK.
For now, AI-generated attacks are still in the minority – little more than 20% of all attacks. But the writing (written by AI) is on the wall, and it is incumbent on enterprises to improve their security defenses now before the tsunami arrives in full.
SecurityWeek talked to Wallin on how enterprises should respond to this growing AI threat. Again, it is nothing new – it must just be done better.
It starts with a basic maxim that is as old as the ancient Delphic Oracle: ‘know thyself’. “Organizations must understand their existing security posture. The key to defending against this increasing volume of attacks and decreasing time to respond is again nothing new. It means focusing on proactive rather than reactive defense. It is security by design from the ground up rather than a bolt-on reaction,” said Wallin.
Proactive cyberdefense is a combination of understanding how attacks work, understanding the enterprise infrastructure and business, and being prepared for all possibilities.
“Zero trust is fundamental to stay in control of credentials, access, and enterprise data. You can no longer build a moat or wall around the infrastructure, so zero trust security by design must be built into the infrastructure, the applications, the API entry points, and the data management,” said Wallin. “This can only be achieved with a full understanding of what is important to the business.”
Interestingly, the father of zero trust, John Kindervag, has published a new book titled Cyber Resilience at Machine Speed: The Zero Trust Model for the AI Era. “Any AI-generated packet still has to move across the same network that attackers have always had to traverse – and correctly implemented zero trust can still halt it,” he told SecurityWeek. He echoes Wallin’s view: the principles of cybersecurity have not been altered by the arrival of AI-driven cyberattacks.
The only thing that has really changed is the statistical likelihood of a breach. This likelihood increases the need for resilience. But if defense starts from the position of ‘know thyself’, the enterprise will understand its key areas. Defensive use of AI can help here. “It is increasingly difficult for everything to be completely free of all unknown vulnerabilities. We can use AI to help us with prioritization, and risk management. We can use it to automate, to discover and to understand, so that we can continuously improve the environments that we’re running as an organization.”
This still doesn’t guarantee that a catastrophic breach won’t happen, so resilience must also include recovery of damaged data to ensure business continuity. This isn’t new, just more pressing – a cyber vault backup facility is necessary. This can be built internally using air-gapping and WORM technology; or acquired externally. The IBM Z mainframe, for example, offers a cyber vault designed for high-end workloads such as banking and core financial services, using automated validation, forensic tools, and immutable backups to test and recover corrupted data rapidly.
AI hasn’t changed the fundamentals of cybersecurity, it has simply made good cybersecurity by design, resilience and recovery more important.
Related: AI Has Changed Attack Speed, Not Security Fundamentals
Related: OpenAI Fires 3 Safety Researchers in Dispute Over AI Risks
Related: Anthropic Fast-Tracks AI Bug Reports to OSS Maintainers, Taps 11 Firms for OT Security
Related: Formula Predicts When AI Chatbots Are at Risk of Turning Bad
Related: Apple to Tighten Full Disk Access Controls in macOS Amid AI Risks
