--- title: What is shadow AI? And why GenAI usage monitoring matters for MSPs and SMDs description: "Generative AI is no longer a side experiment inside businesses. It is moving into normal work: writing, summarizing, coding, research, customer support, etc. That is why the right response to it is not a generic anti-AI message. It is observability first, then governance, then protection." canonical_url: https://www.acronis.com/en/blog/posts/what-is-shadow-ai-and-why-genai-usage-monitoring-matters-for-msps-and-smds/ locale: en last_updated: 2026-10-11 --- [Acronis Blog](https://www.acronis.com/en/blog/) # What is shadow AI? And why GenAI usage monitoring matters for MSPs and SMDs Generative AI is no longer a side experiment inside businesses. It is moving into normal work: writing, summarizing, coding, research, customer support, etc. That is why the right response to it is not a generic anti-AI message. It is observability first, then governance, then protection. [Try Now](https://www.acronis.com/en/products/cloud/trial/) [![Image 2: Acronis Editorial Team](https://staticfiles.acronis.com/images/content/d85707ae91461544212d39836dc32b42.png)](https://www.acronis.com/en/blog/authors/acronis/ "Acronis Editorial Team") [Acronis Editorial Team](https://www.acronis.com/en/blog/authors/acronis/) 10 min read Last updated 09 June 2026 **Author: Alexander Ivanyuk,**Senior Director, Technology Generative AI is no longer a side experiment inside businesses. It is moving into normal work: writing, summarizing, coding, research, customer support, internal search and repeatable workflows. [OpenAI says it](https://openai.com/business/guides-and-resources/the-state-of-enterprise-ai-2025-report/) now serves more than one million business customers, more than seven million ChatGPT workplace seats, and roughly 8x growth in weekly enterprise messages since November 2024. It also reports that weekly users of custom GPTs and projects increased about 19x year to date, which shows that enterprise AI is moving from casual prompts toward persistent, repeatable workflows. That growth matters because the AI problem for most organizations is no longer simply “Should we use AI?” It is “Can we see where AI is already being used, what data it is touching and how much of that use is happening outside approved controls?” That is the real shadow AI question. [IBM reports that 80% of American](https://www.ibm.com/think/insights/rising-ai-adoption-creating-shadow-risks) office workers use AI in their roles, but only 22% rely exclusively on employer-provided tools. In other words, AI adoption is already broad, but official visibility often lags behind actual behavior. ## Shadow AI is broader than public chatbots The most common mistake is to define shadow AI too narrowly. Many people still think it means employees opening ChatGPT in a browser without approval. That is part of the problem, but not the whole problem anymore. Specialists define shadow AI discovery broadly enough to include unsanctioned AI chatbots, AI model-provider APIs, SaaS MCP servers, and AI code generators. Some go even further by describing shadow AI as a mix of SaaS AI apps, GenAI platforms, on-premises AI infrastructure and custom AI agents. That broader definition is important because a company that monitors only a few public chatbot domains can miss a large part of the real exposure. In practice, shadow AI can include public chatbots used through personal accounts, browser extensions, desktop AI assistants, copilots embedded in SaaS applications, developer coding assistants, direct calls to model APIs, local runtimes such as [Ollama](https://ollama.com/), agent frameworks such as [LangChain](https://www.langchain.com/) and custom workflows built on platforms like [Bedrock](https://aws.amazon.com/ai/generative-ai/?trk=1a892402-2c7b-41b5-b0c6-c42f2554f6f0&sc_channel=ps&ef_id=CjwKCAjwhe3OBhABEiwA6392zPbUM-KUCEKE9GF-Xu9s_SGXnSdGawNnoviIAKEtccYdC9TjxjuXNBoCZYgQAvD_BwE:G:s&s_kwcid=AL!4422!3!798628406858!p!!g!!amazon%20generative%20ai!23610841177!198879273572&gad_campaignid=23610841177&gbraid=0AAAAADjHtp_WAEM23qOot093NOl85GDJa&gclid=CjwKCAjwhe3OBhABEiwA6392zPbUM-KUCEKE9GF-Xu9s_SGXnSdGawNnoviIAKEtccYdC9TjxjuXNBoCZYgQAvD_BwE), Azure OpenAI or [Vertex AI](https://cloud.google.com/vertex-ai?utm_source=pmax&utm_medium=display&utm_campaign=Cloud-SS-DR-GCP-1713664-GCP-DR-APAC-SG-en-PMAX-Display-PMAX-Prospecting-Vertex_AI&utm_content=c--x--9196223-17821145463&utm_term&utm_source=PMAX&utm_medium=PMAX&utm_campaign=1011353-Cloud-APAC-SG&utm_content=sg-en&gclsrc=aw.ds&&https://ad.doubleclick.net/ddm/trackclk/N5295.276639.GOOGLEADWORDS/B26943865.344601469;dc_trk_aid=535898303;dc_trk_cid%3D163098484;dc_lat%3D;dc_rdid%3D;tag_for_child_directed_treatment%3D;tfua%3D;ltd%3D&gad_source=1&gad_campaignid=17821146702&gclid=CjwKCAjwhe3OBhABEiwA6392zEU7Cp7xabSZqXp6Qnk05QKLXDdhN876tyKmcM8Y3MJpG-d5DAnp_RoCBVkQAvD_BwE). That is why shadow AI is better understood as unapproved or unmanaged AI use, not just “someone using the wrong chatbot.” ## Why shadow AI appears so quickly Shadow AI does not usually begin with bad intent. It begins with convenience, speed and missing alternatives. Employees use AI because it helps them finish work faster, and if the approved corporate option is missing, weaker, slower or harder to access, they will often route around policy. Workers broadly believe AI improves productivity, and many turn to unauthorized platforms when company-provided tools do not meet their needs. Some analysis makes the same point in more operational language, warning that when organizations ban generative AI outright or move too slowly, employees will find ways to use consumer-grade tools anyway. That is why shadow AI is usually a symptom of two things happening at once: Useful AI capabilities are spreading fast, and governance is arriving later. [Cisco’s 2026 Data and Privacy Benchmark reflects the same imbalance](https://investor.cisco.com/news/news-details/2026/AI-Fuels-Surge-in-Data-Privacy-Investments-and-Redefines-Governance-Cisco-Reports/default.aspx). Cisco says 90% of organizations have expanded their privacy programs because of AI, but 65% still struggle to access high-quality data efficiently, which means the governance side is still catching up to the ambition side. ## Why this becomes a security and governance problem Shadow AI is not only a policy violation issue. It becomes a security problem quickly because AI systems are interactive. People do not just read from them. They type into them, paste sensitive information into prompts, upload files, connect tools and increasingly rely on them inside business processes. That creates several kinds of risk at the same time. Sensitive data can be exposed through prompts or uploads. Employees may use personal rather than managed identities. Business logic can move into unreviewed workflows. Developers may connect AI directly to code, tickets, documents or APIs. And once agents enter the picture, the risk shifts from “What did the model say?” to “What could the system access, decide or do?” NIST’s Generative AI Profile is useful here because it treats generative AI as a distinct risk-management problem, not just as another normal software feature. The financial case is already visible as well. [IBM’s 2025 Cost of a Data Breach](https://www.ibm.com/think/x-force/2025-cost-of-a-data-breach-navigating-ai) work says 63% of organizations in its study had no AI governance policies to manage AI or prevent shadow AI, 97% of breached organizations that had AI-related incidents lacked proper AI access controls, and high levels of shadow AI added an extra $670,000 to the global average breach cost. IBM also says those incidents contributed to broader data compromise and operational disruption. That is a strong argument for treating shadow AI as a real control problem, not as a side topic. ## Personal accounts and unmanaged identities are a major part of the problem One reason shadow AI is difficult to govern is that a large part of it happens outside managed identity. Personal accounts make AI use much harder to inventory, audit and control. [Cyberhaven says](https://www.cyberhaven.com/blog/managing-shadow-ai-best-practices-for-enterprise-security) 32.3% of ChatGPT usage and 24.9% of Gemini usage now occurs through personal accounts. [Netskope reports](https://www.netskope.com/resources/cloud-and-threat-reports/cloud-and-threat-report-2025) that 60% of the enterprise population using SaaS GenAI apps was still using personal SaaS GenAI apps in May 2025, even after many organizations had begun steering users toward approved tools. This matters because “AI use” and “AI use we can govern” are not the same thing. A managed business tenant gives a company more leverage around logging, retention, access control, configuration and policy. Personal accounts weaken all of that. For managed service providers (MSPs), this is one of the most important distinctions to surface early in customer conversations. A customer may believe it has adopted AI safely because it approved one tool, while employees are still doing large parts of their daily AI work elsewhere. ## The shadow AI surface is expanding beyond chat A second mistake is to think the problem is stable. It is not. The shadow AI surface is getting broader. As we already said, AI now includes SaaS apps, GenAI platforms, local AI interfaces and custom AI agents. An average organization is using seven GenAI apps, while organizations in the third quartile are already using in excess of 15, and Netskope is tracking more than 1,550 distinct generative AI SaaS applications. It also reports 8.2GB of data uploaded to SaaS AI apps per organization per month on average, with higher-percentile organizations sending much more. [![Image 3: Acronis GenAI Protection is now live: Secure the AI era](https://staticfiles.acronis.com/images/content/eec37275aced5f0ce36055592787333b.png) Acronis GenAI Protection Acronis GenAI Protection is now live: Secure the AI era Try now](https://www.acronis.com/en/products/cloud/cyber-protect/genai-protection/) Organizations are increasingly using GenAI platforms to build custom applications and agents, that local AI interfaces such as Ollama introduce separate on-premises [security questions](https://www.acronis.com/en/blog/posts/why-genai-brings-new-security-risks-for-msps-and-smds/), and some organizations had users running agents created with popular frameworks on premises. In other words, shadow AI is no longer just “employees chatting with a public model.” It now includes development stacks, platform services, local runtimes and agent frameworks that can sit much closer to enterprise data and systems. ## This is why usage monitoring comes before blocking When organizations first recognize the shadow AI problem, many want to jump straight to blocking. But blocking is rarely the best first move, especially in environments where AI adoption is already widespread and fragmented. Many specialists recommend a dual approach: Provide approved tooling to the workforce and build sufficient observability. Organizations should monitor active models, costs, prompt inputs and outputs and the enforcement of security guardrails, and it recommends endpoint monitoring to detect unauthorized shadow GenAI use. We can make a similar point from the network side, arguing that coaching users toward approved tools is often more effective than flat denial because it reduces unmanaged use without pretending the demand for AI will disappear. That is the right way to frame GenAI usage monitoring. Monitoring is not a weaker form of blocking. It solves a different problem. Blocking stops a known activity. Monitoring reveals an activity pattern that the business still does not understand well enough. Without that visibility, policy tends to be either too late, too broad or aimed at the wrong things. ![Image 4: What is shadow AI? And why GenAI usage monitoring matters for MSPs and SMDs](https://staticfiles.acronis.com/images/content/8dc8daa7b6808e353c42e6d5b3495610.png) ## What GenAI usage monitoring should actually monitor A monitoring program that only counts visits to a few AI websites is incomplete. Modern GenAI usage monitoring needs to function more like observability than like a simple allow / block list. At minimum, it should answer six practical questions. First, what AI applications, APIs, MCP servers, coding assistants, local runtimes and agents exist in the environment? Second, who is using them, through which identities, on which devices and how often? Third, how much data is moving to and from those tools, and which flows look sensitive? Fourth, which of those tools are sanctioned, which are not and who owns them? Fifth, what risk rating or policy status does each tool have? And sixth, what patterns suggest misuse, drift, abnormal cost or growing operational dependency? The ownership point matters more than most teams realize. Organizations need a central registry, access control, visualization, interoperability and security. It also says leaders need to know what agents exist, who owns them, what data they touch and which are sanctioned. That same logic applies more broadly to shadow AI monitoring: Good monitoring is not only discovery, but also inventory, ownership and accountability. ## Sensitive-data visibility now has to be part of the story One of the biggest gaps in older AI discussions is that they talk about “data leakage” in general terms. At the same time data going into GenAI apps includes intellectual property, regulated data, source code and secrets. Cyberhaven says 39.7% of all data movements into AI tools involve sensitive data, including prompts or copy-paste actions, and that 82% of the top 100 most-used GenAI SaaS applications are classified as medium, high or critical risk. Its 2025 report also found that 34.8% of enterprise data shared with AI is sensitive, with source code, R&D materials and sales and marketing data among the common categories. That means usage monitoring has to do more than show that a tool was visited. It should help identify which interactions are likely to involve PII, PHI, source code, secrets, customer data or proprietary business information. Otherwise, the business knows that AI is being used but still does not know where the real exposure is. ## Shadow agents are making the problem more serious The next phase of this problem is shadow agents. Once AI systems start planning steps, using tools and acting on workflows, the visibility requirements become much stricter. Microsoft reports that more than 80% of Fortune 500 companies now use active AI agents built with low-code or no-code tools, and that 29% of employees have turned to unsanctioned AI agents for work tasks. For everyone is very clear that observability comes first: You need to know what agents exist, who owns them, what they touch and how they behave. That is a useful model even for organizations that are not yet deep into agentic deployments, because the transition from “AI that answers” to “AI that acts” changes the risk profile fundamentally. In 2026 coding assistants and AI agents became the “second wave” of workplace AI, and OpenAI’s enterprise report shows that more AI usage is already moving into repeatable, integrated workflows rather than one-off prompts. The practical takeaway is simple: the monitoring problem will get more important as AI becomes more operational. ## What this means for MSPs For MSPs, shadow AI and GenAI usage monitoring should be treated as a managed-environment problem, not just a product feature. The customer question is no longer only “Do we have an AI tool?” It is “How many AI tools, accounts, extensions, APIs, agents and local deployments are already in use across this tenant or customer environment, and which of them are visible, approved and controlled?” That is why MSPs need a layered operating model. First, discover AI use. Second, classify it into sanctioned, tolerated and unsanctioned categories. Third, identify where sensitive-data exposure, unmanaged identity or shadow agents are appearing. Fourth, apply policy, education, redirection and technical controls. Fifth, keep monitoring because the AI surface changes faster than traditional SaaS portfolios do. The current standards and vendor guidance are aligned on this point: Governance is continuous, not one time. NIST’s AI RMF Playbook frames implementation around Govern, Map, Measure and Manage, and Microsoft explicitly treats observability, ownership and cross-functional governance as ongoing disciplines. MSPs also need to explain the economics clearly. If AI logs, prompt events, app discovery and policy outcomes still require manual review for every tenant, the model will not scale well. That is one reason observability, dashboards, risk scoring and reporting matter so much in an MSP context: They are not just security features, but operational requirements. ## What this means for SMDs Small and medium IT deployments (SMDs) should not assume shadow AI is mostly an enterprise problem. In many cases, smaller teams are more exposed because they move faster, have fewer approval layers, and rely on convenience under time pressure. Cyberhaven’s recent work suggests AI adoption is uneven and fragmented, not smooth and centrally managed. In its 2026 view, the top 1% of early-adopter organizations are using more than 300 GenAI tools, while cautious enterprises use fewer than 15. Its 2025 report also found that mid-level employees use AI tools 3.5x more than managers and that workplace AI usage frequency grew 4.6x in one year. The practical meaning for SMDs is that AI usage often becomes real long before governance becomes mature. That is why the right message for SMD customers is not “You need an enterprise AI office.” The right message is simpler: If your people are already using AI in daily work, you need visibility before you can make sensible decisions about approval, data handling, acceptable use and protection. ## Where Acronis GenAI Protection fits For MSPs and SMDs, this is where [Acronis GenAI Protection](https://www.acronis.com/en/products/cloud/cyber-protect/genai-protection/) becomes the practical answer. We describe it as a way to secure generative AI adoption by SMBs by giving partners and customers visibility over GenAI usage and Shadow AI discovery, reducing the risk of users sharing sensitive data with public AI models, and protecting AI interactions from harmful prompts. Acronis GenAI monitoring discovering generative AI applications used across customer environments and providing visibility into adoption patterns and usage trends. ![Image 5: What is shadow AI? And why GenAI usage monitoring matters for MSPs and SMDs](https://staticfiles.acronis.com/images/content/1da2f7c8ed08164542ff239a8f6183fd.png) That matters because it lines up with what the broader market is now demanding: Not only a policy that says “use AI safely,” but an operating model that can actually discover AI use, distinguish approved from unapproved tools, surface risky behavior and add protection as AI becomes more connected to data, tools and agents. ## Conclusion Shadow AI is no longer just a side effect of employees opening public chatbots. It now spans personal accounts, SaaS copilots, model APIs, coding assistants, SaaS MCP servers, local AI runtimes, GenAI platforms and shadow agents. That is why the right response is not a generic anti-AI message. It is observability first, then governance, then protection. For MSPs, GenAI usage monitoring is the first practical step in turning AI from a blind spot into a managed service layer. For SMDs, it is the simplest way to keep using AI productively without losing control over data, identities and business workflows. [Acronis GenAI Protection](https://www.acronis.com/en/blog/posts/acronis-genai-protection-is-now-live-secure-the-ai-era/) answers the real customer need behind shadow AI by connecting visibility, policy, DLP, prompt protection and reporting into a control model that MSPs can actually operate. [![Image 6: Acronis GenAI Protection is now live: Secure the AI era](https://www.acronis.com/en/blog/posts/what-is-shadow-ai-and-why-genai-usage-monitoring-matters-for-msps-and-smds/) Acronis GenAI Protection Acronis GenAI Protection is now live: Secure the AI era Try now](https://www.acronis.com/en/products/cloud/cyber-protect/genai-protection/) [Cybersecurity](https://www.acronis.com/en/blog/tags/cybersecurity/) [Cyber protection](https://www.acronis.com/en/blog/tags/cyber-protection/) [Cyberthreats](https://www.acronis.com/en/blog/tags/cyberthreats/) [MSP cybersecurity](https://www.acronis.com/en/blog/tags/msp-cybersecurity/) [Data protection](https://www.acronis.com/en/blog/tags/data-protection/) [Data loss prevention](https://www.acronis.com/en/blog/tags/data-loss-prevention/) ## Written by ![Image 7: Acronis Editorial Team](https://staticfiles.acronis.com/images/content/d85707ae91461544212d39836dc32b42.png) Acronis Editorial Team Acronis is a global cyber protection company that provides natively integrated cybersecurity, data protection, and endpoint management for managed service providers (MSPs), small and medium businesses (SMBs), and enterprise IT departments. [Previous post](https://www.acronis.com/en/blog/posts/as-compliance-evolves-operational-resilience-becomes-the-real-benchmark/) [As compliance evolves, operational resilience becomes the real benchmark](https://www.acronis.com/en/blog/posts/as-compliance-evolves-operational-resilience-becomes-the-real-benchmark/) [Next post](https://www.acronis.com/en/blog/posts/seagate-and-acronis-partner-with-utah-jazz-to-host-hackathon-to-inspire-the-next-generation-of-innovators/) [Seagate and Acronis partner with Utah Jazz to host hackathon to inspire young innovators](https://www.acronis.com/en/blog/posts/seagate-and-acronis-partner-with-utah-jazz-to-host-hackathon-to-inspire-the-next-generation-of-innovators/) About Acronis Acronis is a leader in AI-native cyber protection, unifying data protection, cybersecurity, IT operations and hybrid infrastructure to deliver business continuity, security and compliance with less cost and complexity. More than 21,000 service providers trust Acronis to protect over 750,000 businesses in 150 countries. Acronis is headquartered in Schaffhausen, Switzerland. [Add as a preferred source on Google](https://google.com/preferences/source?q=acronis.com) ## Related stories Want to learn more? Read the latest stories featuring this solution on our blog. [Acronis Blog](https://www.acronis.com/en/blog/) ![Image 8: Microsoft AI Security for SMBs: Purview Coverage Explained](https://staticfiles.acronis.com/images/blog-card/26d37ae6d05d5d57f26427da530711de.png) [Microsoft AI Security for SMBs: Purview Coverage Explained](https://www.acronis.com/en/blog/posts/microsoft-ai-security-for-small-teams/) See what Microsoft Purview covers for AI security, where additional controls may help, and how Acronis GenAI Protection supports SMBs and MSPs. **September 30, 2026**9 min read ![Image 9: AI data security: protecting sensitive information across AI tools](https://staticfiles.acronis.com/images/blog-card/4018bfcee2b5b2b8fac70b143c616e02.png) [AI data security: protecting sensitive information across AI tools](https://www.acronis.com/en/blog/posts/ai-data-security-protecting-sensitive-information-across-ai-tools/) Learn how SMBs can protect sensitive data in AI prompts, pasted text and file uploads, and how these controls differ from AI security posture management. **September 28, 2026**9 min read ![Image 10: AI security solutions: what they are and how to choose one](https://staticfiles.acronis.com/images/blog-card/d7c45e1cb8fcc53fdb14e5fcd8e0003a.png) [AI security solutions: what they are and how to choose one](https://www.acronis.com/en/blog/posts/ai-security-solutions-what-they-are-and-how-to-choose-one/) Learn how AI security solutions help SMBs manage shadow AI, data leakage and prompt injection. Compare key controls, coverage and MSP delivery. **September 26, 2026**12 min read