The Five Eyes AI advisory assumes a foundation most teams don't have

Aug 4, 20269 min

Jim Armstrong

Sr. Director, Product Marketing, Axonius

In June 2026, the leaders of the Five Eyes cyber security agencies issued a joint cyber advisory:

As the leaders of the Five Eyes cyber security agencies, we are united in our call to action: the evolving landscape of artificial intelligence (AI) is rapidly transforming cyber risk, and we must act swiftly to remain ahead.

This joint warning by the national cyber agencies of the US, UK, Canada, Australia, and New Zealand centers on AI-accelerated cyber threats, especially as frontier AI models compressing the window between discovery of flaws and exploitation. The warning names five practical actions every organization should take. Read individually, each action is reasonable. Read together, they share a quiet assumption: asset intelligence already exists in your organization.

Asset intelligence is a continuously current, cross-domain picture of every asset in an organization's environment and how those assets relate to each other — ownership, network adjacency, identity, and trust. Unlike a static inventory, it reconciles data from every security and IT tool into one decision-grade model rather than leaving each tool as its own isolated system of record.

The Five Eyes advisory's five practical actions each assume this model already exists. For most organizations, that assumption doesn't hold, and it's worth taking the five actions one at a time to see exactly where each breaks when asset intelligence is not in place.

What do you rearchitect, isolate, or decommission first when you can't see your full attack surface?

The first action in the advisory is to “reduce your attack surface,” and there’s only one way you can do that: you must know what makes up your attack surface: which assets are interfacing with your data and processes, and the risks they present. Yet, while many organizations aggregate asset and exposure data in some way, only 45% consolidate it into a single view. And even fragmented consolidation is hard: over two-thirds operate on inventories that are a month or more old.

Unlike a static inventory or single-domain tool, asset intelligence reconciles data across every source to produce a single, continuously current decision-grade model of the environment, including ownership, adjacency, and trust. That model is what makes the Five Eyes practical actions executable: without it, each action is bounded by the limits of whichever tool's view a team happens to be working from.

Axonius was founded to fix these gaps. Soon after Gartner officially recognized the CAASM (Cyber Asset Attack Surface Management) category, consolidating asset data from endpoint, cloud, identity, network, and CMDB tools into one record, Axonius was named as a CAASM vendor in their Hype Cycle for Network Security. Today, with the new, faster-moving threats from AI arriving, asset intelligence as an evolution of CAASM is more critical than ever. Where CAASM consolidated, asset intelligence delivers the decision-grade context and correlation to shrink your attack surface.

Each security and IT tool is a system of record, but only for the piece of the environment they’re intended to manage. The resulting fragmentation still exists in nearly every organization today; however, the Five Eyes advisory's first recommendation assumes that a consolidated view already exists.

What's missing	Where it usually hides
Unmanaged endpoints	Outside primary EDR/MDM coverage in places like network access logs, identity providers, SaaS access logs
Shadow SaaS/shadow AI	Outside procurement and SSO visibility in places like SaaS access logs, expense reports, network and API request logs
OT and regional subsidiary devices	Outside the primary CMDB; often treated as “unmanaged” by IT and Security but tightly controlled by operations teams

When no patch exists, mitigations and compensating controls are the only immediate move.

Patching is great – when a patch exists and you’re able to rapidly apply it. But the threat from AI is that we expect more zero-day vulnerabilities, flaws that are actively exploited before a vendor has shipped a fix. In 2024, attackers weaponized a new vulnerability in roughly 5 days. In 2026, Google Cloud / Mandiant reports that number has gone negative, -7 to be exact. Mitigations and compensating controls are no longer merely fallbacks; they're likely to be the only available immediate actions.

On the defensive side, organizations take a median of 43 days to remediate a known-exploited vulnerability, the ones most likely to get the highest priority, per the latest Verizon DBIR. So how do you decide the best course of action? You need to know your options for mitigation much closer to day 0 than day 43. The answer comes back to decision-grade asset intelligence.

Attack path is the specific route an attacker could take by chaining relationships between assets: identity, network adjacency, trust, privilege. That path also shows you where mitigations or compensating controls can close the same risk faster than waiting for a patch. The challenge here is that while patching options are pretty well-bounded (a vendor releases a patch for the issue), mitigations and compensating controls are not bounded. Your options are a function of how much of the environment's relationship data you can actually see.

Lever	What it is	Bounded or unbounded?	What determines availability
Patch	A fix for the underlying issue	Bounded: one CVE, one patch	Whether the vendor has shipped it yet
Mitigation	An action that closes or narrows a step in an attack path without altering the vulnerability itself; e.g., revoking an unneeded privilege or closing a port	Unbounded	How much of the attack path you can see
Compensating control	A documented, auditable alternative control used when the original control isn't feasible; must meet the intent of the original compliance control they replace (which may or may not affect security posture)	Unbounded	How much of the asset's relationships and exposure you can see; requires formal justification and approval for compliance

Most attack-path tooling only shows part of this picture: cloud-focused tools see cloud relationships, identity/AD-focused tools see on-prem relationships, and external-surface tools see internet-facing relationships. Each is a kind of system of record, bounded by its own domain. The real ceiling on how many mitigation or compensating-control options you can see and consider is comprehensive, cross-domain asset-relationship data, not any single attack-path engine.

What do you do about legacy systems that can never be patched?

The advisory's third action calls for addressing legacy and unsupported systems. The ideal solution is retirement, but that may not be an option. A legacy/unsupported system is, by definition, hardware or software past its vendor's support lifecycle, so "patch faster" will never be an option. How big is the problem? In 2024, IEEE studied network devices from just 3 manufacturers (D-Link, TP-Link, and Netgear) and found 3 million active end-of-life devices.

Legacy systems are rarely missing from every tool. They're visible in one silo or another, or at least as part of your institutional knowledge. But they may very well be invisible in the systems driving a given decision, like the vulnerability scanner that ignores a legacy asset because it alerts on everything and can’t be remediated.

Without the relationship-aware view that asset intelligence delivers, a legacy system with no patch option looks like an open-ended, unaddressable risk. With asset intelligence, it becomes a scoped set of mitigation and compensating-control options like any other asset. The difference is you need to find, document, and maintain those options indefinitely, not just until a patch ships.

Identity and access review depends on the same asset data as vulnerability patching.

Identity systems, including directory services, PAM, and cloud IAM, are frequently reviewed by different teams, on a different cadence, with different tooling than the asset/vulnerability side. But "attack surface," in the advisory's own sense, includes who can reach a system, not just what systems exist. An identity with excess privilege three hops from a crown-jewel asset is part of the same attack path as an unpatched CVE.

Organizations have struggled to achieve a unified view of access and identity posture across fragmented systems, tool sprawl, and poorly normalized identity data. It’s the same asset intelligence problem described earlier, and splitting identity review into a separate process creates yet another silo that breaks the asset intelligence relationship data the rest of this framework depends on. Your identities are part of the attack path for the bad actors; they belong in the same reconciled model as every other asset.

You can't contain an incident along an attack path you've never mapped.

The final guidance in the advisory is to “prepare for incidents before they happen.” Just like every other recommended action before it, testing an incident response plan and containing a breach quickly both depend on knowing what's reachable from the point of compromise. How do you tabletop your way through an attack path you haven’t mapped?

There are two possibilities here:

Run similar AI attack tools to the attackers and see what they find. It’s a bit blind and unbounded, and may take time and be computationally expensive. However, it may also uncover blind spots and gaps. And if you don’t have a great reconciled view of all assets, this may be your best option.
Or, you give the tools an advantage that bad actors don’t have: give them a jump start by feeding them with the context from your asset intelligence. Instead of random chance, they can quickly find the existing toxic combinations that link together to expose your organization.

Only the second option gives your own red teams an advantage that outside attackers lack.

Every Five Eyes action runs through one precondition: asset intelligence.

The Five Eyes advisory is important and directionally correct. The recommendations are also nothing new: these are security fundamentals that have been recommended for a long while now. What is changing is the potential speed from discovery of a flaw to attack, which will push every organization to prepare better, not just react. Every one of the five actions in the advisory runs through the same precondition: asset intelligence, an accurate, relationship-aware view of the environment. The humans in the loop need it to know and prove the risk that exists and how it’s being managed; AI needs it so it doesn’t hallucinate solutions that sound right but do nothing.

To learn more on how you can implement the asset intelligence foundation across your environment, schedule a session with one of our specialists.

FAQ

Does having attack-path visibility from one tool (cloud, on-prem, or external-surface) solve this problem? Only for the domain that tool covers. A cloud-focused attack-path engine won't surface an on-prem identity relationship, and vice versa — the option set you can see is bounded by the asset-relationship data feeding that specific tool, not the full environment.
How fast do organizations actually need to move under this advisory's timeline? The advisory frames the shift as "months, not years." Separately, M-Trends 2026 observed weaponization timelines for new vulnerabilities now show it’s actually -7 days (that’s a negative 7), against typical organizational patch deployment timelines of 43 days for the highest priority vulnerabilities. This is the gap this framework is built to close.
Does this framework replace vulnerability scanning or a CMDB? No, it sits on top of them. Vulnerability scanners and CMDBs each contribute partial asset and exposure data; the framework depends on connecting that data across silos, not replacing the tools that generate it.
What automation is required to protect exposed assets within hours rather than days or weeks? The automation capabilities you need likely already exist. What’s lacking more often is trust. The real answer is data quality: automation stalls because cross-organizational teams don't trust the asset data and business context driving it, and the fix is asset intelligence accurate enough to act on without manual validation.
How can we use AI to enhance predictive detection and behavioral monitoring? AI-driven detection and monitoring is only as good as the asset context it reasons over. You don’t need a prediction based on general stats or industry trends from six months ago; you need your company’s specific context and trends. Behavioral anomalies without ownership and relationship data produce alerts, not actions.
Our cloud security tool already shows us an attack path; do we really need another tool? Asset intelligence isn’t another tool that sits alongside your cloud security tools, or any other security tool, for that matter. It’s the reconciliation layer that sits above the other tools and creates the unified intelligence and context necessary for the entire organization to respond to threats at AI machine speed.