Artificial intelligence is becoming part of everyday business operations, from customer support and content creation to workflow automation and AI agents. But as more AI tools connect to the internet, databases, business applications, and user accounts, they also introduce new cybersecurity risks that organizations must manage.
A new report from internet intelligence company Censys highlights the scale of this change. Its 2026 State of the Internet Report, released on October 14, 2026, found that internet-facing AI infrastructure grew by 63% over a nine-month period, reaching nearly 268,000 systems.
The report also found that 76% of n8n hosts with identifiable software versions fell within the affected range of at least one recent critical vulnerability. This finding does not mean that every exposed host was compromised, but it highlights why AI-related infrastructure needs the same careful security management as other business software.
For website owners, developers, and businesses adopting AI, the message is straightforward: adding AI to a workflow should also mean reviewing what that workflow can access, how it is protected, and whether it is exposed to the public internet.
What Are the New AI Security Risks?
AI security risks are threats that arise when AI applications, automation platforms, models, or connected services are exposed or inadequately protected. These risks can include unauthorized access, exploitation of vulnerable software, leaked credentials, and unintended access to connected business systems.
Censys reported a 63% increase in internet-facing AI infrastructure over nine months in 2026. The findings highlight the need to identify exposed AI tools, update vulnerable software, restrict access, and monitor connected systems.
However, an internet-facing service is not automatically vulnerable or compromised. The actual risk depends on its configuration, software version, permissions, and security controls.
What Did the Censys 2026 Report Find?
Censys monitors internet infrastructure to identify systems and services that are accessible online. Its latest report examined how this infrastructure is changing as AI adoption expands.
Here are the main findings relevant to AI security.
| Finding | What it means |
|---|---|
| 63% growth in exposed AI infrastructure | More AI-related systems became accessible from the public internet during the tracked nine-month period. |
| Nearly 268,000 systems | Censys identified this number of internet-facing systems running 44 AI-related products. |
| Approximately 113,926 exposed n8n hosts | n8n, a workflow automation and AI orchestration platform, was the most widely exposed product identified in the report. |
| 76% of n8n hosts with identifiable versions | These hosts were running software versions within the affected range of at least one recent critical vulnerability. |
| Approximately 54,000 OpenClaw hosts | OpenClaw was another widely observed AI-related product in the exposed infrastructure. |
Source: Censys 2026 State of the Internet Report and its October 14, 2026 announcement.
These numbers describe infrastructure observed by Censys, not every AI system in existence. They also do not establish that every identified host was exploitable.
Nevertheless, the findings point to a growing challenge: businesses may be deploying AI-connected tools faster than they can inventory, update, and secure them.
Why Are Internet-Exposed AI Tools a Security Concern?
The risk is not limited to an AI chatbot answering questions. Many AI applications operate behind the scenes, connecting models to software, data, accounts, and automated workflows.
Consider a business that uses an AI workflow to summarize customer enquiries, update a spreadsheet, retrieve information from a database, and send notifications to a team.
The AI model is only one part of that arrangement. The workflow platform, connected accounts, API credentials, database permissions, and hosting environment also need protection.
If an attacker gains unauthorized access to a poorly secured component, the potential consequences may extend to other services connected to it.
1. AI automation can connect multiple business systems
Workflow automation platforms such as n8n allow users to connect applications and build automated processes. When these workflows incorporate AI, they may also interact with databases, customer-management systems, file storage, and other services.
This integration can save time, but it makes permissions especially important.
If a workflow has more access than it needs, a compromised account or vulnerable component could create a wider security problem than the original application alone.
What businesses should do: Review each workflow’s permissions and remove access to systems or data that it does not need.
2. Vulnerable software can create entry points
AI tools are still software, and they can contain security flaws just like other applications.
The Censys findings concerning n8n illustrate why software maintenance matters. Among the observed n8n hosts whose versions could be identified, 76% were running versions affected by at least one recent critical vulnerability.
That statistic should not be interpreted as proof that 76% of all n8n installations—or all AI tools—are vulnerable. It applies to the specific observed group described in the report.
Even so, running outdated software can leave an organization exposed to known security issues.
What businesses should do: Keep AI platforms, plugins, integrations, dependencies, and hosting software updated. Review security advisories and verify that installed versions are not affected by relevant vulnerabilities.
3. Poorly protected credentials can increase the impact
AI applications often require credentials to access other services. These may include API keys, application tokens, database credentials, or account permissions.
If credentials are exposed or misused, the consequences may extend beyond the AI tool itself.
For example, an automated system with permission to read private files should not automatically receive permission to delete them, modify financial records, or administer an entire account.
What businesses should do: Store secrets securely, rotate credentials when exposure is suspected, and grant each application only the permissions necessary for its assigned tasks.
4. Unknown AI deployments can go unnoticed
Not every AI tool is deployed through an organization’s formal IT process. Employees and teams may experiment with new services, connect applications, or host their own tools without a complete security review.
This can create what security teams call shadow IT or shadow AI: technology being used without sufficient organizational visibility or oversight.
The problem is not experimentation itself. It is losing track of which systems exist, what information they can access, and who is responsible for securing them.
What businesses should do: Maintain an inventory of approved AI services, locally hosted models, automation platforms, integrations, and internet-facing endpoints. Establish a straightforward process for reviewing new tools before they handle sensitive information.
What Does This Mean for Website Owners?
You do not need to run a large technology company to benefit from the report’s lessons. Website owners also rely on software, plugins, APIs, hosting accounts, and external services that may introduce security risks.
For example, a website might use AI to generate content drafts, answer customer questions, summarize submissions, or automate administrative tasks. Each additional integration can create another component that needs to be managed.
Here are five practical priorities.
Keep WordPress, themes, and plugins updated
For WordPress site owners, outdated core software, themes, and plugins can create avoidable security risks.
Install updates from trusted sources, remove unused plugins and themes, and check whether installed components have known security advisories. Back up the site before major updates and verify that important features still work afterward.
Protect administrator accounts
Use unique passwords, enable two-factor authentication where available, and limit administrator access to people who genuinely need it.
Review old accounts and revoke access when it is no longer required. An AI integration should not automatically receive the same privileges as a website administrator.
Review AI plugins and third-party integrations
Before connecting an AI service to your website, understand what information it receives, which permissions it requests, and where data is processed.
Avoid granting broad access simply because it makes setup easier. If a plugin only needs to read a specific set of content, investigate whether narrower permissions are available.
Protect API keys and private information
Do not publish secret keys in front-end code, public repositories, or pages accessible to visitors. Use appropriate server-side secret management and restrict keys to the services and actions they require.
Also review whether the AI service needs access to customer information, unpublished content, or other sensitive records in the first place.
Monitor the systems that support your website
Keep an eye on security alerts, administrator activity, unusual API usage, failed login attempts, and unexpected changes to website files or configurations.
Use reputable security and hosting tools suited to your setup. If your website connects to AI automation services, include those services in your security review instead of monitoring the website alone.
An AI Security Checklist for 2026
Use this checklist when introducing AI tools or reviewing an existing setup.
- Inventory: Identify AI services, automation platforms, plugins, APIs, and hosted models in use.
- Exposure: Confirm which services need to be reachable from the public internet and restrict unnecessary access.
- Updates: Check software versions and apply relevant security updates.
- Permissions: Give each AI tool and workflow only the access it needs.
- Credentials: Store API keys and tokens securely and review who can use them.
- Data protection: Avoid sending sensitive information to services that do not need it.
- Monitoring: Review logs and alerts for unusual activity.
- Backups: Maintain reliable backups and periodically verify that recovery works.
- Human oversight: Require approval for consequential actions, such as deleting important data or changing critical business settings.
- Incident response: Know how to disable a risky integration, revoke credentials, and investigate a suspected compromise.
This checklist is a practical starting point, not a guarantee of security. Organizations handling sensitive data or critical services may need a more formal security assessment.
How AI Agent Security Fits Into the Bigger Picture
The growth of exposed AI infrastructure is part of a wider shift toward AI systems that can do more than generate text.
AI agents and automated workflows may interact with tools, retrieve information, and perform actions across connected services. Those capabilities can be useful, but they also make access controls and oversight more important.
A system that can only answer a question has a different risk profile from one that can change a record, access a private database, or trigger an external action.
This is why AI safety cannot rely entirely on a model following instructions. Security also depends on the surrounding infrastructure: permissions, authentication, network restrictions, logging, and clear limits on what the system is allowed to do.
Prompt instructions remain useful, but a prompt is not a substitute for access control. If an action should never be permitted, enforce that restriction in the application or system rather than relying only on a written instruction to the AI.
For a deeper look at this issue, read Promptenza’s related coverage:
- AI Agents Are Becoming More Autonomous: What Recent Security Incidents Teach Us
- AI Worms Explained: The Security Risks of Self-Replicating AI
- Claude Bypassed Website Restrictions: What Anthropic’s AI Agent Tests Reveal
These topics connect to the same core lesson: as AI systems gain access to more tools and services, the security of those connections becomes increasingly important.
Does This Mean AI Tools Are Unsafe?
No. The Censys findings do not establish that AI tools are inherently unsafe, nor do they show that every exposed system has been compromised.
Internet accessibility is only one part of a security assessment. The level of risk depends on factors such as software versions, authentication, configuration, network controls, permissions, and the sensitivity of connected systems.
AI tools can provide real benefits when deployed responsibly. The goal is not to avoid AI altogether, but to adopt it with the same care used for other software that handles business data or performs important tasks.
Organizations should assess individual deployments instead of treating all AI products as equally risky.
The Bottom Line: Secure AI From the Start
Censys’s 2026 State of the Internet Report offers a timely reminder that AI adoption is expanding beyond visible chatbots into the infrastructure connecting applications, data, and business processes.
The reported 63% growth in internet-facing AI infrastructure shows how quickly this part of the technology landscape is changing. The findings about exposed n8n hosts also demonstrate why software maintenance and accurate visibility matter.
For website owners and businesses, the most useful response is practical: know which AI systems you use, understand what they can access, keep them updated, protect their credentials, and monitor the services they connect to.
AI security should be part of deployment—not an afterthought once a problem appears.
As AI tools become more deeply integrated into everyday work, the organizations best prepared to benefit will be those that combine useful automation with strong, well-maintained security controls.
Frequently Asked Questions
What are the biggest AI security risks in 2026?
Key risks include exposed AI infrastructure, vulnerable software, poorly protected credentials, excessive permissions, unmonitored integrations, and unauthorized access to connected business systems. The specific risks depend on how each AI application is deployed and what it can access.
What did Censys find about exposed AI tools in 2026?
Censys reported that internet-facing AI infrastructure grew by 63% over nine months, reaching nearly 268,000 systems running 44 AI-related products. It also found that 76% of observed n8n hosts with identifiable versions were affected by at least one recent critical vulnerability.
Does an exposed AI tool mean it has been hacked?
No. Internet exposure means a system is accessible from the public internet; it does not, by itself, prove that the system is vulnerable or compromised. Security depends on factors such as authentication, configuration, software versions, and access controls.
Why is n8n mentioned in the Censys report?
n8n is a workflow automation platform that can connect applications and services, including AI-powered workflows. Censys identified approximately 113,926 exposed n8n hosts, making it the most widely exposed AI-related product in the report. The findings highlight the importance of securing exposed automation infrastructure.
How can website owners reduce AI security risks?
Website owners can update WordPress and its plugins, enable stronger account protection, review AI integrations, restrict permissions, protect API keys, monitor activity, and maintain reliable backups. They should also avoid giving AI tools unnecessary access to private website data.
Are AI agents more dangerous than ordinary AI chatbots?
Not automatically. However, an AI agent that can use tools, access databases, or make changes to connected services may have a wider range of potential consequences if its permissions or surrounding systems are poorly secured. Its risk depends on the actions it can perform and the controls governing those actions.
Can prompt engineering prevent AI security incidents?
Good prompting can clarify instructions, constraints, and expected behavior, but it cannot replace technical security controls. Authentication, permissions, network restrictions, monitoring, and approval requirements are needed to enforce limits on what a system can actually do.
What is the first step in securing AI infrastructure?
Start by identifying the AI applications, automation tools, integrations, and internet-facing services your organization uses. Once you know what exists and what each system can access, you can prioritize updates, restrict unnecessary exposure, and address the most important risks.
Sources and Further Reading
- Censys — 2026 State of the Internet Report
- Censys announcement — Exposed AI Infrastructure Grew 63% in Nine Months
- Axios — AI’s Growing Internet Footprint
