When innovative technologies emerge, organizations typically move through the same cycle. A clear promise comes to the fore, competitive pressure accelerates decision-making, and security tries to adapt but often gets left behind. A good example of this is public cloud.
Cloud adoption promised scalability, speed and offered access to capabilities people could not build on premise. But its lack of definition created uncertainty as well as opportunity. In particular, larger enterprises struggled against competitors and often faced significant challenges, such as shadow IT schemes deployed outside the central IT departments, and losing competitive agility.
As a result, IT teams faced ambiguity, and their security postures were reactive and on the backfoot. AI is now repeating that cycle at a greater scale. It is unfolding in successive waves, each bringing greater capabilities and deeper integration. As AI continues to grow, failure to understand and respond to those waves properly is one of the biggest risks that businesses face today.
Risk across AI maturity
The first AI wave saw enterprises focus on predictive analytics. Predictive models sat behind the scenes, with capabilities such as data lakes, large-scale pattern recognition, and machine learning. As a result, these capabilities rarely attracted executive scrutiny. For security teams, the issue lied in data protection; their focus was to protect sensitive information from misuse.
That changed with the rise of generative AI. Tools that could draft content, write software, and create imagery pushed AI into the everyday workflow. But that attention masked a deeper issue. AI and Generative AI were merged into one concept, concealing meaningful distinctions in risk. Defensive efforts were focused on containment and monitoring. According to research from Zscaler, titled The Ripple Effect: A Hallmark of Resilient Cybersecurity’ report, seven in ten organizations lack clear visibility into shadow AI usage, while 56% believe sensitive data is already being exposed through unsanctioned tools. In response, organizations default to extending existing security tools rather than redesigning their security approach. But agentic AI marks the turning point across the threat landscape.
Agentic AI and autonomous execution
Nowadays, agentic AI is making headlines, and for reasons leaders can’t afford to ignore. Unlike other AI tools, it can operate without human supervision across the enterprise. For organizations, agentic AI offers multiple benefits, including its ability to act autonomously like a “digital teammate” and execute complex tasks. The resilience report finds that 34% of enterprises are in the deployment phase of agentic AI, while 42% of organizations are currently assessing it. What remains concerning is that governance and security protections are still missing in nearly half of AI deployments. That absence exposes the limitations of legacy security thinking.
Not all AI systems produce the same kind of risk. Agentic AI introduces behavior and system integrity risks. In contrast, generative and predictive AI shifts the risk to data exchange. When firms enable AI agents to interact directly with ERP environments or financial systems, the consequences of compromise escalate. Today, agentic AI is following a similar trajectory to the one that we saw during cloud adoption. However, this time, automation unlocks attack paths faster than security teams can reasonably adapt.
The gap between control and exposure
In response to the emergence of agentic AI and its threats, enterprises are moving quickly to reinforce their defenses by allocating more budget to cybersecurity. The vast majority (90%) of organizations increased their cyber-resilience spending in 2025, and 96% updated their resilience strategy. Yet, the research exposes a gap. Despite this increase in spending, 61% concede that their strategies are still largely inward-looking. Consequently, this creates a false sense of assurance for enterprises as they believe that they’re in control of everything inside of their organizations. This raises a critical question about whether firms are prepared for what lies beyond, such as the ecosystem of external partners, platforms, and AI-driven supply chains.
The danger escalates when an agentic AI’s autonomous capabilities extend beyond organizational boundaries, especially as agentic AI becomes more involved in driving supply chain automation. Industries such as retail, logistics, and manufacturing are already moving in this direction as they pursue optimization and sustainability goals. When agentic AI begins coordinating work between companies, threats propagate faster than any single organization can contain without resilience engineered upfront. Instead of having isolated security incidents, we’ll see these breaches ripple outwards throughout the supply chain.
Adapting to a changing threat landscape
Defending against this agentic-AI enabled threat landscape doesn’t mean ignoring existing security principles. However, to keep up with advancing threats these principles must evolve. In fact, agentic AI requires the same Zero Trust controls used to manage people, such as issuing identities, adopting least-privileged access, and strict oversight. Where the difference lies, however, is that these guardrails must be applied at a much greater scale. When an agent operates outside of those parameters, its actions should be immediately flagged and treated with the same suspicion as human behavior.
Under these conditions, segmentation is non-negotiable. Equally, organizations must understand that AI-security must not be viewed as an add-on. The Ripple Effect research revealed a crucial finding, that 52% of leaders believe that their security systems fall short against advanced threats. With countless organizations already struggling to manage sophisticated threats, agentic AI amplifies weaknesses that have existed for years. What is clear is that this widespread lack of enterprise preparedness in response to the current threat landscape raises acute questions about organization’s broader readiness to tackle emerging threats such as agentic AI and quantum. readiness for agentic AI and quantum computing.
The current discussion about Mythos is another good example of new, expanding frontiers. It underscores the speed, opportunities, and perils of AI and demonstrates that it has never been more important to adopt the core architectural principles of Zero Trust. With such an approach, IT security professionals have actually had the tools at hand for a number of years to support the removal of an organisation’s attack surface, reduce lateral movement, and stop data exfiltration. They simply need to accelerate adoption and can therefore reduce risk by going dark and securing transactions with policies brokered by a security platform.
A shift in approach
One clear takeaway from both the cloud era and the ongoing AI evolution is that reactive security will always eventually reach its limit. The speed of technological change now outpaces governance and regulatory processes, leaving enterprises exposed if they wait for guidance to catch up, or for incidents to define priorities.
This reality calls for a different mindset. As AI advances, controls must evolve alongside it. Agentic systems are becoming increasingly autonomous, interconnected, and more tightly woven into daily operations. For organizations, history shows that standing still only leads to repeating the mistakes of the past.
