TL;DR: Critical exposures are often hiding in AI infrastructure, mobile-connected ICS assets and services on non-standard ports, while older software and permissive hosting choices remain persistently exposed, according to Censys’ 2026 State of the Internet Report. The practical lesson is that discovery gaps, not just vulnerability severity, continue to define real-world internet risk.
At a glance
What this is: Censys argues that internet exposure is being missed in places many security programs still under-scan, especially AI infrastructure, ICS systems on mobile networks and services running off default ports.
Why it matters: IAM, NHI and broader security teams need to treat discovery scope as a control issue, because hidden infrastructure can expose credentials, privileged access paths and unmanaged services before conventional monitoring notices.
By the numbers:
- Between October 2025 and July 2026, the number of hosts exposing AI-related tools grew more than 63%.
Context
Internet exposure management starts with a simple problem: if defenders do not look in the right places, assets stay invisible even when they are already exposed. Censys’ report uses that lens across AI, ICS, vulnerability response, threat infrastructure and the broader internet, showing that visibility gaps remain a governance issue, not just a tooling issue.
The article is especially relevant to identity security because some of the most exposed systems are not the user-facing applications people notice, but the infrastructure that stores credentials, orchestrates workflows and connects to external services. That makes discovery, ownership and access scoping part of the same control conversation as monitoring and remediation.
Key questions
Q: How should security teams manage unknown internet-facing assets?
A: Security teams should treat unknown internet-facing assets as governance exceptions, not just missing inventory. The first step is to create a repeatable discovery process that attributes ownership, then route high-risk findings into an enforced remediation workflow. If nobody can own the asset, the organisation should assume the exposure is active until proven otherwise.
Q: Why do exposed AI infrastructure components create more risk than the visible application layer?
A: Because the support stack often holds the credentials, connectors and execution paths that let an attacker move beyond the interface. If those components are exposed, compromised or poorly scoped, the issue becomes identity and access control as much as model security. The application may be the visible surface, but the supporting infrastructure usually determines the blast radius.
Q: What breaks when teams assume services only run on default ports?
A: They miss a meaningful share of exposed systems, especially web services, proxies and remote access tools configured on non-standard ports. That creates a blind spot in scanning, alerting and prioritisation, because exposure exists even when the service is not listening where the tool expects it to be.
Q: How should organisations prioritise forgotten or unchanged systems in exposure management?
A: Treat them as likely ownership and lifecycle failures, not just patching backlog. If a system has stayed exposed for a long time with little version change, it may be abandoned, unmonitored or outside normal remediation workflows. Those conditions deserve priority because they often indicate a wider governance gap rather than a single missing fix.
Technical breakdown
Why AI infrastructure exposure is harder to see than application risk
AI exposure is often discussed as a model problem, but the real exposure surface can sit underneath the model in workflow orchestration, service connectors and credential stores. These components hold the access needed to move between systems, which means compromise can extend beyond the AI interface itself. The report’s point is not that every exposed AI tool is exploitable, but that visibility into the supporting stack is now part of the control plane for AI risk. See the AI Infrastructure Workload Identity Guide.
Practical implication: Map AI support systems, not just AI apps, into the same asset and identity inventory.
Why mobile-connected ICS assets evade normal discovery
Operational technology frequently lives outside the network assumptions that enterprise tools use for scanning and ownership. When ICS devices sit on consumer or mobile networks, they can be exposed to the internet while remaining invisible to the team responsible for them. That creates a classic governance failure: the asset exists, the exposure exists, but the organization has no reliable way to identify it fast enough to remediate or notify the right owner.
Practical implication: Extend asset discovery beyond corporate address space to include externally reachable operational systems.
Why default-port assumptions create blind spots in exposure monitoring
Many scanners and inventories implicitly assume common services live on their standard ports, but attackers do not need to follow that convention. When SSH, HTTP(S) or proxy services move to non-standard ports, the service is still exposed, but the detection logic may miss it. This is a visibility problem in the same family as shadow infrastructure: the risk exists because the monitoring model is too narrow for how services are actually deployed.
Practical implication: Tune exposure monitoring to hunt by service behaviour, not only by default port.
Threat narrative
Attacker objective: Exploit unseen or underestimated internet-facing systems to gain access, persistence or reach into connected environments.
- Entry begins with exposed internet-facing infrastructure that is not covered by normal discovery or ownership processes.
- Escalation occurs when that infrastructure provides access to credentials, workflows or management interfaces that extend beyond the initial asset.
- Impact follows when attackers abuse the overlooked system to reach connected services, maintain access or target downstream environments.
Breaches seen in the wild
- Gravity SMTP CVE-2026-4020 API Keys Exposure: CVE-2026-4020 in Gravity SMTP exposes API keys via single HTTP request across 100,000 WordPress sites.
- Meta Muse agent hijack 2026: An undocumented Muse setting let local malware hijack Meta's personal AI agent, steal its authentication material and abuse user access.
Read and download The State of NHI & AI Agent Breach Report 2026, covering 200+ breaches impacting Non-Human Identities including AI Agents.
NHI Mgmt Group analysis
Visibility gaps are now a core governance failure, not an inventory nuisance. The report shows that organisations keep underestimating exposures because their tooling starts from assumptions about where assets should live and how they should look. That is a dangerous model when internet-facing infrastructure can shift quickly, hide on unexpected networks, or expose services outside the normal monitoring path. The practical conclusion is that discovery quality has become a first-class control.
Administrative AI infrastructure is becoming an identity problem as much as an application problem. When platforms connect agents, store credentials and broker access to external services, they function like high-value identity fabric for machine activity. That means the security question is not just whether the AI tool is vulnerable, but whether the credentials and service relationships around it are visible, scoped and governed. Treating those components as workload identities behind AI platforms changes the risk picture materially.
Non-standard exposure patterns undermine assumptions that many security programmes still rely on. Defenders often model services, ports and ownership as if infrastructure were neatly categorized, but the article shows the internet does not cooperate with that model. Exposure management has to follow what is reachable and what is authenticated, not what was expected during design. That is a better fit for modern infrastructure and a better basis for prioritisation.
Old software that never changes is a governance signal, not just a technical nuisance. Stable exposure over long periods usually means something has been abandoned, forgotten or left outside the operating model. That is exactly the sort of condition where identity and access sprawl tend to accumulate, because nobody is actively maintaining the surrounding control plane. Practitioners should treat persistent old versions as evidence of weak lifecycle governance, not merely stale patching.
What this signals
Exposure management is becoming a control issue for AI, not just a scanning issue. When workflow platforms and service connectors sit outside the normal inventory model, the programme has to govern the supporting infrastructure as tightly as the application itself. That is where identity, credential and privilege risk becomes operationally relevant, because access is often embedded in the plumbing rather than the interface.
Discovery models need to follow reachable services, not assumed architecture. Non-standard ports, mobile-connected systems and forgotten internet-facing hosts all undermine the idea that exposure can be found by standard patterns alone. The next maturity step is to align exposure detection with what is actually reachable and what can actually be authenticated, not what the network diagram says should exist.
For practitioners
- Expand discovery beyond standard network assumptions Include mobile networks, non-standard ports and external service exposure in routine asset discovery so the programme does not miss systems that live outside corporate address space.
- Inventory AI support systems as governed assets Track workflow orchestration, service connectors and credential stores alongside the AI application itself, because those supporting components often hold the real blast radius.
- Validate exposure on all ports, not just defaults Use scan and detection logic that identifies services by behavior and banner data rather than assuming SSH, HTTP and proxy services will appear only on standard ports.
- Prioritise long-lived, unchanged assets for review Flag software and systems that have remained exposed with little change over long periods, since they often indicate forgotten infrastructure and weak ownership.
Key takeaways
- The report’s central warning is that exposure is being missed in infrastructure layers many programmes still do not inspect closely enough.
- Its examples span AI support systems, ICS assets on mobile networks and services running on non-standard ports, showing how visibility gaps create risk.
- The practical response is to widen discovery, strengthen ownership of exposed assets and align monitoring to real service behavior rather than assumptions.
Related NHI Mgmt Group resources: AI Infrastructure Workload Identity Guide · Zero-Day and Exploited Vulnerabilities Guide · API Key Management Guide
Standards & Framework Alignment
This section maps relevant standards and security frameworks to the operational risks and controls described in this guidance.
MITRE ATT&CK addresses the attack and risk surface, while NIST CSF 2.0, CIS Controls v8 and CSA Cloud Controls Matrix set the governance and control requirements practitioners need to meet.
| Framework | Control / Reference | Relevance |
|---|---|---|
| NIST CSF 2.0 | ID.AM-01 — Assets are inventoried and owned | The article is fundamentally about assets that remain exposed because discovery and ownership are incomplete. |
| DE.CM-01 — Networks and network services are monitored to find potentially adverse events | The report shows monitoring gaps where exposed services exist outside default assumptions. | |
| Recommendation — Maintain a complete asset inventory that covers externally exposed systems, including mobile-connected and non-standard assets. Monitor networks and services broadly enough to detect internet exposure on non-standard paths. | ||
| CIS Controls v8 | CIS-1 — Inventory and Control of Enterprise Assets | The report’s blind spots are first and foremost inventory failures across internet-facing infrastructure. |
| Recommendation — Continuously inventory internet-facing assets and reconcile them against observed exposure. | ||
| MITRE ATT&CK | TA0007;TA0006 — Discovery; Credential Access | The article describes exposed infrastructure and credential-bearing AI support systems that enable attacker discovery and access. |
| Recommendation — Map exposed services and credential-bearing infrastructure to discovery and credential access techniques for detection planning. | ||
| CSA Cloud Controls Matrix | IAM — Identity and Access Management | AI support infrastructure in the article stores credentials and brokers external access, making IAM governance directly relevant. |
| Recommendation — Apply cloud IAM controls to the AI support stack, not only to the visible application layer. | ||
Key terms
- Attack Surface Discovery: The process of finding and classifying assets that can be reached, tested, or abused by an attacker. In modern AppSec, discovery must be continuous because build pipelines, AI-assisted code, and microservice sprawl can change the attack surface faster than manual review can track.
- Non-Standard Port: A port number that is not the default port commonly used by a service. For Rsync, choosing a non-standard port can reduce routine scanning and lower casual interference, but it does not replace authentication, firewall controls, or encryption. Security gains come from reducing exposure, not from obscurity alone.
- Workload Identity: The identity assigned to a software workload, such as a containerised application, serverless function, or microservice, enabling it to authenticate to other services without storing static credentials.
- Shadow Infrastructure: Infrastructure that is provisioned outside the central tracking process and therefore escapes normal governance. It often appears when teams need to move quickly, but it creates hidden identity exposure because the credentials, roles, and exceptions attached to it are easy to miss.
Deepen your knowledge
NHI Foundation Level course, the industry's only accredited NHI security programme, covers NHI governance, workload identity and secrets management in the context of modern access risk. It helps security practitioners build the control foundations needed to govern machine identities alongside human identity programmes.
Published by the NHIMG editorial team on October 22, 2026.
NHI Mgmt Group, the independent authority on Non-Human Identity, IAM, and Agentic AI security. nhimg.org