Five Eyes Alliance Warns We Have Months, Not Years. Is Your Infrastructure Secure by Design?
The cybersecurity agencies of the Five Eyes alliance have issued a direct warning to business leaders: artificial intelligence is rapidly transforming cyber risk, and the time available to prepare is shrinking.
According to the joint statement from agencies in the United States, United Kingdom, Australia, Canada and New Zealand, frontier AI models are expected to fundamentally change both offensive and defensive cyber capabilities. Their timeline is stark: “not years, it is months.”
The immediate concern is easy to understand: AI lowers the technical barriers for malicious actors and allows them to identify vulnerabilities, develop attacks, and operate at greater speed and scale. However, focusing only on AI-powered attackers misses the other half of the risk.
Organizations must also look inward at the AI systems already connected to their applications, data, and infrastructure and ask whether those systems were built with the governance and security controls needed to withstand what is coming.
AI is changing both sides of the threat equation
AI is increasing the speed and sophistication of cyberattacks. It is also becoming an active participant inside the enterprise.
Organizations are no longer using AI only to generate content or answer questions. AI systems can increasingly access sensitive information, interact with applications, write and execute code, provision cloud resources and make changes across production environments.
That creates a dual-sided risk:
Attackers can use AI to find and exploit weaknesses faster.
An organization’s own AI systems can introduce or amplify risk.
A powerful model does not have to be inherently malicious to become dangerous. It may be manipulated, compromised, or instructed to take an action its designers did not anticipate. If that model or an agent has excessive access to infrastructure or sensitive systems, or agency to make decisions and execute autonomously, the impact can extend well beyond the model itself.
Every organization integrating AI therefore inherits a critical responsibility: understanding not only what the AI can do, but what it can reach and what it is authorized to change.
Three areas demand immediate attention
The Five Eyes agencies call on organizations to reduce their attack surfaces, accelerate patching, address legacy systems, strengthen identity and access controls, and prepare for incidents before they occur. They also make secure-by-design and secure-by-default practices central to their recommendations.
For organizations building and operating cloud and AI infrastructure, that guidance should begin with three areas.
1. Publicly exposed infrastructure
Organizations must know which resources are exposed to the internet, whether that exposure is necessary, and how quickly unintended exposure can be identified and corrected.
In a dynamic cloud environment, this cannot be answered through a periodic inventory or annual assessment. Resources are continuously created, changed, and removed. A secure configuration can drift, and a single infrastructure change can unintentionally create a new entry point.
Reducing the attack surface requires continuous visibility and controls that prevent unnecessary exposure before it reaches production.
2. Exploitable vulnerabilities and misconfigurations
As AI improves vulnerability discovery and exploit development, organizations have less time to move from awareness to remediation.
Patching faster is part of the answer, but it is not the whole answer. Security teams also need to know which vulnerabilities exist, where the affected assets are running, how exposed and reachable they are and whether mitigating controls are in place.
That context is often fragmented across cloud platforms, scanners, ticketing systems and infrastructure tools. The resulting manual work slows the response precisely when speed matters the most.
Organizations need a continuous connection between infrastructure visibility, risk prioritization, and remediation.
3. AI with excessive agency
An AI agent’s risk is determined not only by the intelligence of its underlying model, but also by the authority granted to it.
Can it access sensitive information? Change network configurations? Create identities or modify permissions? Take decisions about a patient’s healthcare? Delete resources? Shut down water supply without approval?
The more consequential its potential actions, the more important its operating boundaries become.
AI agents should be governed through least-privilege access, explicit policy constraints, continuous validation and complete auditability. High-impact actions should require the appropriate level of human approval.
An autonomous system should never receive autonomous trust.
Secure-by-design must begin before provisioning
Traditional cloud security often follows a familiar pattern: infrastructure is created, a security tool scans it, an issue is discovered and an engineer is asked to fix it.
That approach leaves a window between deployment and remediation. As the cyber clock accelerates, that window becomes increasingly difficult to accept.
The stronger approach is to apply security and governance before infrastructure is provisioned.
Architecture and configuration should be evaluated against organizational policies at the point of creation. Noncompliant infrastructure should be corrected or blocked before it becomes an exposed production resource. Once deployed, it should be continuously monitored for drift and changing risk.
This turns secure-by-design from a stated principle into an enforceable operating model.
It also allows organizations to increase infrastructure velocity without asking security teams to manually review every change. Developers receive approved paths to deployment, security teams maintain consistent control and leaders gain evidence that requirements are being applied across the environment.
Defenders must enable AI usage, but protect the enterprise
The Five Eyes agencies are not arguing that organizations should retreat from AI. AI usage will only increase. Companies are racing to demonstrate innovation, gain competitive edge, increase margins and revenue. So AI usage is inevitable. The argument is about safe usage, preparing not just for an incident as security teams have traditionally done but also for reducing the attack surface in the first place. Not just identify misconfigurations and vulnerabilities earlier, but also improve software quality, design systems and infrastructure with security baked in, enable security by default.
Here’s the rationale. Industry estimates suggest that every enterprise carries more than 3000 misconfigurations at any given point in time. For decades, we have been detecting more and remediating less. Secure by design is a way to ensure these 3000 misconfigurations don’t exist in the first place. The urgency of this approach comes from the fact that AI is revealing new vulnerabilities at an unprecedented pace, 49,000 to date in 2026 already surpassing the entirety of 2024 or years before. Patching at this pace is impossible, especially if patches are not even available. Secure by design plugs the entry points and closes the paths. That reduces the risk that the vulnerabilities will be exploited.
AI deployments are rapidly increasing the attack surface and the complexity of the attack surface. Following traditional models of building and deploying AI systems is a disaster waiting to happen. We are going to see an increase in the number of misconfigurations driven by the race and pace of AI adoption. Enabling innovation while protecting the enterprise necessitates rethinking security, using first principles. That leads us and that is what led the Five Eyes Alliance to secure by design, secure by default, securing before you implement a new system.
This is a whole-of-organization leadership responsibility
Security teams have always known that secure by design is the right way. What they have struggled with is how to do it without adding friction and enabling business. Shift left with scanning code has created friction with the engineering teams. Shift left with scanning infrastructure code has created friction with the infrastructure teams and does little to protect infrastructure built or changed outside this process. It has only slowed down the infrastructure deployments and teams under immense go to market pressures have found shortcuts. Expectations of pace have only dramatically increased with AI causing more slop and mistakes.
Intelligent automation that empowers the infrastructure teams with security expertise they have to otherwise rely on from the architects and security teams on to provide and automation that takes the guesswork and gruntwork out of the coding and deployment process, needs a redesign of the traditional process and that needs a leadership commitment. CIOs and CTOs that own the infrastructure must drive this change to keep business innovating and resilient. CISOs have to be the ones evangelizing and initiating these discussions with them.
The Five Eyes statement makes clear that cyber resilience is not solely an IT or security problem. It is a business risk tied to operational continuity, market confidence and long-term value. That means the Boards and executives need to start asking harder risk questions in addition to costs and ROI:
Do we know what infrastructure and AI systems are operating across our environments?
Can we identify unnecessary exposure and critical vulnerabilities continuously?
Are security requirements enforced before deployment or assessed afterward?
How do we know what decisions are delegated to the AI agents, what controls, audit trails and evals do we have in place for the efficacy, reliability of those decisions, and our controls over the decisions, execution and data access?
Can we contain an unsafe action before it creates wider damage?
The window to prepare is short, skills are scarce
We are at an inflection point. Organizations must prepare for both dimensions of that reality: external actors using AI against them and internal AI systems operating with access to their environments increasing the attack surface. AI skills are scarce and the skills required to secure AI that is constantly and rapidly evolving are even rarer to find.
Your enterprise and business is not slowing down the adoption of AI. Security teams and infrastructure teams need to collaborate to implement a solution that embeds a system of continuous, enforceable control and rigor in the deployment processes that support business innovation and competitive advantage at the pace of AI.
The cyber clock has changed. The window to get ahead is open, but it is short.
Five Eyes warns that every exposed resource and overprivileged AI agent can increase the risk and requires security by design to be implemented in months.
If you are finding that AI security expertise capable of keeping pace with this shift is in short supply and the gap is widening faster then you are not alone. InviGrid closes that gap in the platform itself, embedding secure-by-design expertise directly into how infrastructure gets provisioned. Contact us: info@invigrid.com or Contact Us using this form.
