A version of this article was published on MarketWatch.com.

Amazon recently revealed it faces close to 1 billion cyber threats every day. That staggering number underscores a dramatic shift: Cybersecurity threats are no longer occasional, isolated incidents but a relentless, daily battle.

This wasn’t always the case. In the late 1980s, I wrote the first lines of cybersecurity code for personal computers. Back then, the digital landscape was relatively simple. Cyber threats were fewer, less sophisticated, slower and targeted a narrower scope.

Over the years, the explosion of digital infrastructure — from the web to email, cloud computing, and now generative AI — has exponentially increased the attack surface. What once required “red phone” discussions at the highest levels of government and industry now happens routinely — often without the victims realizing they’ve been compromised.

Between then and now, I have had a front-row seat in fighting bad actors. I’ve written early code for PCs, led Symantec as CEO and, later, advised top U.S. government officials, boards of directors and CEOs on how to understand the current threat landscape.

We’re at another inflection point today. Generative AI (Gen AI) is not only changing how people interact with technology but also reshaping the threat landscape.

Gen AI will completely transform the cyber battlefield

The rise of Gen AI has democratized access to powerful technology, making it easier than ever to create, distribute and exploit vulnerabilities. In 2025, I expect the emergence of AI agents — autonomous programs acting on behalf of users — will mirror the transformational shift we saw when email became mainstream in the ’90s.

With every wave of technological innovation comes an equally potent wave of threats. Gen AI and AI agents are no exception. For security executives and startups, this means anticipating entirely new categories of attacks — such as adversarial AI, data poisoning and agent impersonation — and introducing solutions before they overwhelm security teams.

In the 1990s, internet users tolerated pop-ups reminding them to update their antivirus software. By the 2010s, with the advent of automatic updates, the industry shifted to operate seamlessly in the background. Today, with notification fatigue at an all-time high, any solution that disrupts user flow risks immediate rejection. According to a recent survey, 78% of millennials have deleted an app because of notification or alert fatigue.

The challenge is compounded by Gen AI. AI agents will make decisions autonomously, introducing new layers of complexity. For example, a compromised agent could execute unauthorized actions across an enterprise system faster than traditional defenses can respond.

Security providers must design solutions that require minimal user intervention. Success will depend on platforms that integrate into existing systems, limit noise and prioritize usability without compromising efficacy. User-dependent processes will only hinder adoption.

Effective security requires operational efficiency

Cybersecurity advancements often come with trade-offs. For instance, improvements in threat detection have increased false positives, creating inefficiencies for security teams. Filtering too aggressively risks blocking legitimate traffic, while too lenient an approach leaves vulnerabilities exposed.

This balancing act will become even more critical with AI agents. Imagine an AI-powered email filter blocking essential business communications or a defensive AI agent taking overly restrictive actions, disrupting enterprise workflows.

Solutions must prioritize operational efficiency alongside security efficacy. Security executives and startups must understand that the “cure” cannot create more problems than the original threat. Cybersecurity platforms should provide actionable intelligence, clear prioritization and adaptable defense mechanisms tailored to the unique needs of enterprises.

According to the International Monetary Fund, the economic impact of cyberattacks could exceed $23 trillion by 2027, up from about $8.4 trillion in 2022. As AI agents become ubiquitous, the frequency, sophistication and potential consequences of attacks will escalate.

Throughout my time in cybersecurity, I’ve seen transformational changes in technology, security team approaches and attacks. At the center are technological shifts, such as the rise of email and internet usage, which dramatically increase the need for security software. Along the way, specific viruses, like Code Red, Melissa, Michelangelo and I Love You, played significant roles in raising awareness and driving the adoption of security solutions.

Another significant attack, Stuxnet, proved that sophisticated cyber attacks can take down or cripple critical infrastructure along with digital networks. In that case, the attack disrupted Iran’s uranium enrichment program and expanded the global cyberattack surface area into physical operations.

Nowadays, attacks on the scale of Stuxnet, Code Red and Melissa happen regularly. These cyberattacks have the potential to topple networks, permanently disable systems and make a real-world impact. What happens next depends on how humans respond.