How to contain shadow AI, as it takes on multiple forms and spreads throughout your environment

Shadow AI used to be confined to a few simple scenarios, such as AI chatbots and AI browser extensions. Today, it spans a broad spectrum of complex vectors, making it harder to detect and control. In the latest installment of our Cybersecurity Awareness Month blog series, learn how employee security training, AI usage governance, and exposure management can help you tackle shadow AI successfully.

Key takeaways

1. Integrating shadow AI into security awareness training helps employees recognize its risks and learn how to safely adopt AI tools. Security teams should foster open communication with employees and try to understand why they use unsanctioned AI products.
2. Shadow AI now spans local models, AI coding assistants, autonomous agents, and many other asset types across your environment, making it a pervasive and complex problem.
3. You can reduce shadow AI incidence by enacting an AI acceptable use policy that provides clear guidelines and a list of approved AI tools, and by addressing shadow AI as part of your unified exposure management program.

It’s an ordinary workday at a typical mid-to-large enterprise, and the following is happening without the knowledge of the IT and security departments:

• A marketing manager installs an unvetted, open-source Model Context Protocol (MCP) server to secretly bridge his desktop AI assistant with the company’s internal customer database.
• A sales executive downloads a free AI browser extension to summarize a web page, unknowingly granting a third party read-access to her confidential CRM dashboards.
• An accounting supervisor creates custom AI chatbot skills to automate his reporting, feeding unredacted financial projections into a public generative AI tool.
• An engineer bypasses software restrictions to download and run an unvetted, locally hosted AI model, creating a massive blind spot for endpoint security.
• A business analyst hands over her broad administrative credentials to a consumer-grade AI agent so it can autonomously crawl the company’s private data warehouse.
• A developer pastes proprietary, unreleased source code into a free, web-based AI assistant for quick debugging, inadvertently surrendering corporate IP to a public model’s training set.

Yes, we’ve come a long way from the days when instances of shadow AI consisted mostly of employees oversharing work information with their personal AI chatbot. With AI now permeating enterprises’ operations, shadow AI has spread and morphed accordingly across the attack surface. It has become riskier and harder to detect, especially as autonomous AI agents proliferate. 

For example, security incidents involving organizations’ shadow AI assets have more than doubled to 43%, compared with 20% last year, according to IBM’s “2026 Cost of a Data Breach Report.” Meanwhile, the average data breach cost of shadow AI-related incidents hit $5.39 million, about $400,000 above the global average.

IBM's report found a direct impact on data privacy, operations, and compliance:

• Almost half of shadow AI security incidents resulted in data loss or data compromise
• 42% caused operational disruption
• 21% triggered regulatory fines

During a 30-day period earlier this year, the Tenable One Exposure Management Platform found 457 million security issues across 7,000 customer environments using 274 detection plugins built specifically for AI. That averaged out to about 62,000 AI-related exposures per customer.

“The enterprise AI landscape of 2026 bears little resemblance to the AI landscape of even three years ago,” the Cloud Security Alliance (CSA) wrote in its April report, “The Invisible Enterprise: Shadow AI and the Ungoverned Frontier.”

Because AI adoption has “dramatically outpaced” organizations’ AI governance, many enterprises now have a dangerous “invisible layer of AI infrastructure” that hasn’t been inventoried, and thus isn’t subjected to policy enforcement, access controls, or audits, according to the CSA report.

In this third installment of Tenable’s weekly Cybersecurity Awareness Month blog series, we’ll unpack best practices organizations can adopt to tackle shadow AI in its multiple forms — AI code assistants, models, applications, APIs, embedded features, agents, MCP servers, training data, browser extensions, and more — and mitigate its risks, such as compliance failures, system breaches, and data leaks.

3 steps to help you mitigate the risk from shadow AI

1. Boost employee and leadership awareness and education

A logical place to start your campaign against shadow AI is by teaching employees and leaders how to recognize and prevent the issue.

Employees

Just last month, the U.K. government stressed how important it is for organizations to address the human factor in shadow AI, specifically the disconnect between employees’ fast adoption and organizations’ lag in approving AI tools and setting AI usage policies.

For example, 34% of employees polled in Thomson Reuters’ “Future of Professionals 2026” report admitted to knowingly and deliberately using AI wares their organization hasn’t approved. 

Why? “They are frustrated by the quality of sanctioned tools or the lack of a clear AI strategy,” reads a Thomson Reuters article about the report, published in June. 

The U.K. National Cyber Security Centre (NCSC) recommends focusing on three key elements: 

1. Understand why staffers use unapproved AI tools.
2. Establish a “positive cybersecurity culture” marked by open communication between employees and the security team.
3. Securely integrate AI systems into the workplace.

“By raising awareness of the risks of shadow AI use within your organization and understanding the needs of employees, you can help them get the benefits of new technologies while using them securely,” an NCSC cloud researcher wrote in the blog “The hidden risks of shadow AI.” 

In other words, focus on employee education, not punishment, and include shadow AI in your security awareness program training. 

Business unit leaders

Proactively reach out to business unit leaders to learn about the AI tools they’re using, their reasons for using them, and the benefits they yield. Engaging in this needs-assessment exercise can help you decide which AI tools to allow and which ones to prohibit.

2. Establish a clear AI governance policy

Avoid being seen as an enemy of AI adoption. Extremely restrictive approaches, such as “block by default” policies, tend to backfire because they drive employees and teams to push their AI usage further underground, especially if these tools boost their productivity.

Instead, you should adopt an AI acceptable use policy that outlines clear rules for who can use AI, which tools they can use, and what data they can input, in a way that doesn’t compromise security and compliance. 

Your AI policy should also clarify levels of accountability by outlining the responsibilities of employees, managers, and the AI governance team. For example, it should stress that employees must verify the accuracy of AI-generated content, because each individual is ultimately responsible for the quality of their work.

The comprehensive policy should clearly outline AI usage guardrails, such as rules for data handling, detail the formal process and criteria that the security and IT teams will use for approving and adopting AI tools, and provide a curated list of authorized tools.

It should further help you assess whether you should outright block a discovered shadow AI tool, or whether you can allow it with the appropriate controls. 

Some questions to ask:

• Does the AI vendor claim ownership of what your users input into its tool?
• Does the tool encrypt data in transit and at rest?
• What are the vendor’s vulnerability and incident disclosure processes?
• Does the product comply with regulations relevant to your organization, like HIPAA?
• Does the vendor use your data to train public AI models?
• In which geographic region does the vendor house your data?
• Which types of industry compliance does the AI vendor adhere to? 

Finally, the AI usage policy should be a living document your organization continually revises, as AI technology advances and the organization’s usage evolves.

“Leaders should focus on creating a governance framework that balances innovation with risk management, ensuring employees feel empowered to use GenAI responsibly,” reads Deloitte’s “GenAI Workforce Survey,” which was published in September. The report found that 31% of U.K. employees use AI tools without their organizations’ permission.

3. Address shadow AI as part of your broader exposure management program

With AI now pervasive across most enterprises, shadow AI is present throughout your IT environment in a wide variety of manifestations. This is precisely the type of complex and multidimensional security issue that a unified exposure management platform helps untangle. 

At Tenable, we believe that trying to detect and control shadow AI as a standalone problem with isolated point tools is the wrong approach.

To successfully mitigate AI security risks, including shadow AI, you need to gather contextually rich AI security data, so you can correlate and analyze it alongside your other exposure data. That way, you can see how your security issues combine to create attack paths that lead to your most critical systems and data.

These are some key ways in which an exposure management platform can help you address shadow AI risk.

• Obtain comprehensive visibility of your AI environment via a continuously updated inventory of your AI assets – what they are, who and what can access them, and which data they can access – wherever they are in your environment including in your:
  • Data centers
  • Endpoints
  • Web apps
  • Cloud workloads
  • Network traffic
  • Application development pipelines
  • Enterprise AI platforms
  • Operational technology (OT) systems
  • IoT infrastructure
• Detect shadow AI assets’ security flaws, including vulnerabilities, misconfigurations, excessive permissions, insecure output handling, and exposed training data. 
• Contextually map and prioritize the security issues that require immediate attention across your entire attack surface based on factors like AI asset criticality and business impact. 
• Take action and automate remediation, including by leveraging agentic AI capabilities, to carry out the fixes needed to close your shadow AI exposures.
• Reduce the risk of data getting shared improperly via AI prompts and uploads by: 
  • Discovering and classifying sensitive data
  • Auditing AI assets’ permissions and enforcing least-privilege policies
  • Monitoring data flows in real time
  • Detecting and blocking sensitive intellectual property, proprietary source code, and customers’ personally identifiable information (PII) before exposure or compromise

Check back next week when we’ll finish our Cybersecurity Awareness Month series with a blog about the cybersecurity challenges of state and local governments. Read the first two blogs in the series:

• How to mitigate the risk from AI-generated apps built by your '‘citizen coder’ employees
• Critical infrastructure orgs face heightened cyber risk from nation-state groups, AI threats

To learn more about shadow AI and how Tenable can help you address this issue, check out the following resources:

• What is shadow AI? (guide)
• Introducing Tenable One AI Exposure: A new standard for securing AI usage at scale (blog)
• Security for AI: A strategic framework for closing the AI exposure gap (blog)
• Mastering AI security risks with unified exposure management (on-demand webinar)
• Tenable Hexa AI: Automating exposure remediation with agentic routines (blog)
• What is AI security posture management (AI-SPM)? (guide)
• What is an AI acceptable use policy (AUP)? (guide)