Shadow IT Statistics 2026: Shadow AI, SaaS and Breach Costs

Shadow IT statistics for 2026 describe a visibility problem that AI has made more expensive. Gartner expected 75% of employees to acquire, modify or create technology outside IT's visibility by 2027, up from 41% in 2022. IBM's 2026 Cost of a Data Breach Report found that security incidents involving shadow AI more than doubled to 43%, from 20% a year earlier, and that those incidents averaged USD 5.39 million. Verizon's 2026 DBIR counts 45% of employees as regular AI users on corporate devices. Below are the verified numbers, what each measures, and how to find your own shadow assets.

Shadow IT statistics at a glance

Every figure below was checked against the publisher's own page or report on 4 November 2026. The studies measure different things, so read the "What it measures" column before comparing two rows.

Primary sources for each row are linked in the sections below and listed under Sources and References.

Three kinds of evidence sit in this table, and they should not be blended. IBM measures organizations that already suffered a breach, so its percentages describe breached companies, not all companies. Verizon and Netskope report telemetry from security products, which shows real behavior but only inside the customers those products cover. Gartner, Cisco and Microsoft report surveys and forecasts. A forecast is a planning assumption, not a measurement, and a survey answer is a self-report. None of these numbers can be added together into one "shadow IT rate."

What counts as shadow IT and shadow AI

Shadow IT is any hardware, software, cloud service or account that people use for work without the knowledge or approval of the team responsible for securing it. Typical examples are a SaaS tool paid for on a corporate card, a marketing site launched by an agency on a company subdomain, a test database left reachable from the internet, or a personal file-sharing account used for work documents.

Shadow AI is the same pattern applied to AI. IBM's 2026 report defines it as workers using unapproved AI. In practice it covers employees pasting data into consumer chatbots, signing in to AI services with personal accounts, connecting AI plug-ins or agents to company data without review, and teams deploying models or AI APIs outside the approved inventory.

Shadow IT is not always reckless. Gartner frames it as a shift of technology work into business units, and most shadow tools are adopted to get work done faster. The security problem is narrower: an asset nobody inventories is an asset nobody patches, monitors, tests or offboards. That is why the statistics below focus on visibility, cost and exposure rather than on how often employees break a policy.

How widespread is shadow IT?

Employees and technology outside IT's view

Gartner's March 2023 cybersecurity predictions put the scale plainly: "By 2027, 75% of employees will acquire, modify or create technology outside IT's visibility – up from 41% in 2022." The Gartner press release presents this as a strategic planning assumption. The 41% is a 2022 survey-based estimate and the 75% is a forecast; Gartner's underlying research is behind a paywall, so this page quotes only the public release.

SaaS bought outside IT

SaaS purchasing data tells the same story from the finance side. Zylo's 2026 SaaS Management Index, which the vendor says is built on more than 40 million SaaS licenses and USD 75 billion in spend under management, reports that business units now control 81% of SaaS spend while IT directly manages just 15%. Expense-based SaaS spend rose 267% year over year, and ChatGPT became the most expensed application (Zylo announcement, 29 January 2026). Zylo also reports that large enterprises add an average of 21 applications per month. These are a SaaS management vendor's customer figures, not a census of all companies, but they show why an inventory taken once a year goes stale.

Personal cloud apps and unmanaged devices

Netskope's Cloud and Threat Report 2026, based on anonymized data from its platform between 1 October 2024 and 31 October 2025, found that 31% of users in the average organization upload data to personal cloud apps every month, and that 60% of insider threat incidents involve personal cloud app instances (Netskope Cloud and Threat Report 2026). Regulated data made up 54% of the data policy violations linked to personal cloud apps, followed by intellectual property (22%), source code (15%) and passwords and API keys (8%).

Unmanaged devices are the hardware side of shadow IT. In the Verizon 2025 DBIR executive summary, Verizon's analysis of infostealer credential logs found that 46% of compromised systems with corporate logins in their data were non-managed and held both personal and business credentials. Verizon attributes these most likely to bring-your-own-device programs or company devices used outside policy. Credentials stolen from a device that IT does not manage are invisible to endpoint controls until someone uses them. Our insider threat statistics cover the non-malicious insider side of the same problem.

What shadow IT costs when a breach happens

Poor application visibility is a top cost factor

IBM's 2026 Cost of a Data Breach Report analyzed 30 factors that raise or lower breach costs against the USD 4.99 million global average. "Lack of visibility into the number and location of applications (shadow IT)" added USD 201,165 to the average breach. Only supply chain breaches (USD 227,250) and security system complexity (USD 208,265) added more, with noncompliance with regulations close behind at USD 201,112 (IBM Cost of a Data Breach Report 2026). IBM measures each factor in isolation, so the factors cannot be summed into a single figure.

The same analysis shows the other side of the ledger. Offensive security testing, which IBM defines as red teaming, penetration testing or vulnerability testing, was associated with USD 211,339 lower breach costs. In IBM's 2025 edition, attack surface management tools were associated with USD 160,547 lower costs. These are associations in a breach sample, not proof that any one control causes a saving, but they point the same way: finding and testing what you own pays.

For the full cost picture across industries and regions, see our analysis of the cost of a data breach.

Shadow data and scattered storage

Shadow IT usually means shadow data. IBM's 2024 report found that more than one-third of breaches involved shadow data, meaning data stored in unmanaged sources, and that 40% of breaches involved data stored across multiple environments (IBM newsroom, 30 July 2024). In the 2026 report, breached data stored across multiple locations again took the longest to identify and contain, at 256 days, an improvement from 277 days the year before. Data that sits in an unknown SaaS tenant or a forgotten bucket is harder to find during an investigation as well as before one. Our cloud security statistics break down how misconfiguration and identity drive those cloud exposures.

Shadow AI statistics

Shadow AI statistics now dominate shadow IT research, because AI tools spread faster than any earlier category of unsanctioned software. This section stays on one question: how unsanctioned AI use creates exposure. For AI adoption, model attacks and AI testing data more broadly, see our AI security statistics. Our AI security guide explains where shadow AI fits in an AI security program.

How many employees use AI outside approved channels

• 78% bring their own AI. The Microsoft and LinkedIn 2024 Work Trend Index found that 75% of knowledge workers use generative AI at work and 78% of AI users bring their own AI tools to work, rising to 80% at small and medium-sized companies. The survey covered 31,000 workers in 31 markets between 15 February and 28 March 2024. The same study found 52% of people who use AI at work are reluctant to admit using it for their most important tasks, which is exactly the behavior that keeps AI use out of sight.
• 45% are regular AI users on corporate devices. Verizon's 2026 DBIR reports that 45% of employees are now regular users of AI, authorized or not, on corporate devices, up from 15% the previous year (Verizon 2026 DBIR executive summary).
• 67% use non-corporate accounts. In the same data, 67% of users accessed AI services with non-corporate accounts on corporate devices, a slight decrease from the year before. Verizon's 2025 report put the split at 72% non-corporate email identifiers and 17% corporate emails without integrated authentication, which Verizon said most likely suggests use outside corporate policy.
• 69% of security leaders see it. A Gartner survey of 302 cybersecurity leaders, run from March to May 2025, found 69% of organizations suspect or have evidence that employees use prohibited public GenAI (Gartner press release, 19 November 2025).

Can security teams see shadow AI?

Most organizations say they cannot see it well. In the Cisco 2025 Cybersecurity Readiness Index, a double-blind survey of 8,000 private sector security and business leaders in 30 markets, 60% said they lack confidence in their ability to identify unapproved AI tools in their environments. Smaller businesses struggled most (68%, compared with 51% of medium and large enterprises). Sixty percent said they cannot see the specific prompts or requests employees make to GenAI tools, and 22% of companies said employees have unrestricted access to publicly available GenAI tools.

Governance is moving backward among breached organizations. IBM's 2026 report found 68% of breached organizations lacked AI governance to manage AI or detect shadow AI, up from 63% in 2025. Only 32% had policies in place, down from 37%. Among the governance measures organizations reported, regular audits for unsanctioned AI fell to 29% from 34%, and only 19% said their governance and security teams coordinate.

What data leaves through shadow AI

Verizon's 2026 DBIR found that shadow AI became the third most common non-malicious insider action in its data loss prevention (DLP) dataset in 2025, a fourfold increase in share from the previous year. Source code was the data type most often submitted to external GenAI models, followed by images and other structured data. In 3.2% of DLP policy violations, research and technical documentation was uploaded to unauthorized AI systems. The figure is based on 858,440 untrusted DLP events that targeted GenAI tools.

Source code is a particular exposure for software companies. Code pasted into a consumer AI tool can carry hostnames, internal API routes, configuration values and occasionally credentials. Once it leaves, the organization loses control of where it is stored and how long it is kept, and the leak usually leaves no trace in the systems the security team monitors.

Shadow AI breach statistics: frequency, cost and impact

IBM's breach data is the clearest measure of what shadow AI costs once something goes wrong.

Three points stand out. First, the share of incidents involving shadow AI more than doubled in a single study cycle. IBM's 2025 report described its 20% as the share of studied organizations that reported a breach involving shadow AI, with a further 11% unsure; IBM's 2026 report describes the 43% as the share of security incidents involving shadow AI. Second, shadow AI incidents cost about USD 400,000 more than the 2026 global average (USD 5.39 million against USD 4.99 million; our calculation from IBM's figures). Third, about one in five shadow AI incidents in 2026 ended with a regulatory fine, a question IBM asked for the first time.

The 2025 edition adds detail the 2026 edition does not repeat. Organizations with high levels of shadow AI saw average breach costs of USD 4.74 million, USD 670,000 more than organizations with low levels or none (USD 4.07 million). In shadow AI incidents, customer personal data was compromised 65% of the time, against 53% across all breaches, and intellectual property 40% of the time, against 33% overall (IBM Cost of a Data Breach Report 2025, PDF).

Shadow AI forecasts

Forecasts belong in a separate column from measurements. Gartner predicts that by 2030 more than 40% of enterprises will experience security or compliance incidents linked to unauthorized shadow AI, and recommends enterprise-wide AI usage policies, regular audits for shadow AI activity and GenAI risk evaluation in SaaS assessments. IBM's measured 43% describes breached organizations in 2025 and early 2026, so it is not evidence that Gartner's forecast has already been met.

From shadow AI to shadow agents and machine identities

The next layer of shadow AI is not a chatbot tab. It is an AI agent, plug-in or connector that someone wires into company systems with its own credentials. IBM's 2026 report found that fewer than half of organizations (46%) secure non-human identities in AI workflows, and only 40% use access controls on AI models and data. Among organizations that reported a breach of their AI models or applications, 92% lacked proper AI access controls, and the most common root causes were compromised APIs, applications or plug-ins and cloud misconfigurations affecting AI workloads.

An unapproved agent with a long-lived API key is shadow IT with the ability to act. If you are deploying agents or connecting them through the Model Context Protocol, our guides to agentic AI security and MCP security explain how to test what an agent can reach and do. For measured hijacking rates and how many organizations have no defined limits on agent access, see our AI agent security statistics.

Where shadow IT shows up on your attack surface

Shadow IT matters to an attacker only when it is reachable. Most shadow assets leave public traces: DNS records, certificates, login pages, exposed storage and new API endpoints. The table below maps each statistic to the external signal it usually produces and to what a tester should validate. It is DeepStrike's method guidance, not data from the cited studies.

Two of these rows depend on external discovery. Our attack surface management guide explains how external discovery finds hosts that are missing from internal inventories and why each finding still needs validation. Domains deserve their own watch list: teams register campaign domains and forget them, and attackers register lookalikes of yours. Our guide to homoglyph attacks shows how lookalike domains are built and monitored.

US federal agencies already work to a discovery cadence. CISA's Binding Operational Directive 23-01, issued 3 October 2022, required federal civilian agencies by 3 April 2023 to run automated asset discovery every 7 days and to start vulnerability enumeration across all discovered assets every 14 days. The directive excludes ephemeral assets such as containers and third-party-managed SaaS, which is a useful reminder that IP-based discovery alone will not find SaaS shadow IT. Private companies are not bound by the directive, but its weekly cadence is a reasonable benchmark.

What the numbers mean: a 30-day shadow IT baseline

The statistics point to one practical conclusion: you cannot secure assets you have not found, and you cannot trust an inventory that is checked once a year. This is a starting sequence that a security team can run in about a month. It is DeepStrike's method guidance, not a requirement from any standard.

1. Build the external inventory first (week 1). Enumerate domains, subdomains, certificates, cloud IP ranges and exposed services from the outside. Output: a list of every internet-facing asset with a named owner or an "unknown owner" flag.
2. Pull the money trail (week 1). Ask finance for expensed and card-paid software over the last 12 months. Zylo's data suggests this is where most unsanctioned SaaS and AI tools appear first. Output: SaaS and AI tools missing from the approved list.
3. Check identity coverage (week 2). Compare SaaS and AI tools against your single sign-on and MFA coverage. Tools with local passwords or personal accounts are the highest-risk shadow assets. Output: a list of tools to move behind SSO or retire.
4. Map AI data paths (week 2). Review DLP, proxy or browser logs for uploads to AI services, and list AI connectors and API keys with access to company data. Output: approved AI tools, tools to block, and integrations to rescope.
5. Search for exposed credentials (week 3). Check infostealer logs and breach data for corporate credentials, especially from unmanaged devices. Output: forced resets and session revocations.
6. Test what you found (weeks 3 to 4). Prioritize unknown-owner and internet-facing assets for manual testing, including authentication, authorization and exposed data. Output: validated findings with severity, not just a scanner list.
7. Set a cadence (week 4). Repeat external discovery at least weekly, in line with the CISA BOD 23-01 benchmark, and test new assets and new API endpoints as they appear. Output: a recurring report that shows new assets, owners and open findings.

How we compiled these shadow IT statistics

Selection. We included a statistic only if we could read it on the original publisher's page or report and record its date, sample and method. We prioritized independent breach research (IBM and Ponemon Institute, Verizon), analyst firms (Gartner), large surveys by technology companies (Microsoft, Cisco) and US government guidance (CISA). We also used two vendor datasets with published scope (Netskope telemetry and the Zylo SaaS Management Index) and labeled them as vendor data. We excluded figures that appear only in vendor blogs without a traceable original study, figures attributed to Gartner that we could not find in a public Gartner release, and paywalled analyst research we could not read.

Report years. Report titles carry the publication year, not the year of the events. IBM's 2026 report covers breaches from March 2025 to February 2026, and Verizon's 2026 DBIR covers incidents from October 2024 to November 2025. Netskope's 2026 report covers October 2024 to October 2025.

Calculations. The only derived figure on this page is the USD 400,000 difference between IBM's 2026 shadow AI breach cost (USD 5.39 million) and its 2026 global average (USD 4.99 million). All other numbers are quoted as published.

Limits. IBM's percentages describe breached organizations, not all organizations. Telemetry reflects the customers of the vendor that collected it. Surveys reflect what respondents say. None of these sources gives a single, representative US shadow IT rate, and no such rate is claimed here.

Frequently asked questions

What percentage of employees use shadow IT?

Gartner estimated that 41% of employees acquired, modified or created technology outside IT's visibility in 2022 and forecast 75% by 2027. For AI specifically, Microsoft's 2024 Work Trend Index found 78% of AI users bring their own AI tools to work, and Verizon's 2026 DBIR counts 45% of employees as regular AI users on corporate devices.

How much does shadow IT add to the cost of a data breach?

IBM's 2026 Cost of a Data Breach Report found that a lack of visibility into the number and location of applications (shadow IT) added USD 201,165 to the USD 4.99 million average breach cost. Breaches involving shadow AI averaged USD 5.39 million in the same report.

What are the key shadow AI statistics for 2026?

IBM's 2026 report found security incidents involving shadow AI rose to 43% from 20%, at an average cost of USD 5.39 million, and 21% of those incidents led to a regulatory fine. Verizon's 2026 DBIR found 67% of users access AI services with non-corporate accounts on corporate devices. Gartner forecasts that more than 40% of enterprises will have a shadow AI incident by 2030.

Is shadow AI a bigger risk than shadow IT?

Shadow AI is a subset of shadow IT, but it moves data faster. A personal file-sharing account stores data; an AI tool can also process it and return it in new forms, often under a personal account. IBM's data shows shadow AI breaches cost more than the global average, and Verizon found source code is the data type most often sent to unauthorized AI tools.

How do you detect shadow IT?

Combine three views: external discovery of domains, subdomains, certificates and exposed services; finance data on expensed and card-paid software; and identity, DLP or proxy logs that show which SaaS and AI services people actually use. Then validate the assets you find with testing, because discovery alone does not show whether an asset is exploitable.

Can a penetration test find shadow IT?

A penetration test can find shadow IT within its scope, especially when it starts with external reconnaissance and includes newly discovered hosts and APIs. A one-time test only shows a point in time. Continuous discovery paired with continuous testing catches assets that appear between scheduled tests.

Is shadow IT a compliance issue?

It can be. Data stored or processed in an unapproved tool may sit outside the controls that privacy and security regulations expect, and IBM's 2026 report found that about one in five shadow AI incidents ended with a regulatory fine. Whether a specific tool breaks a specific rule depends on the data involved and the jurisdiction.

Next steps

The numbers agree on one thing: shadow IT is now normal, and shadow AI is growing faster than the governance around it. The defensible response is not a ban but a cycle of discovery, ownership and testing that runs as often as your environment changes.

DeepStrike's continuous penetration testing service watches public JavaScript files, API documentation, release changelogs and GraphQL schema changes, then has testers assess new features and endpoints as they ship. The Premium plan on our penetration testing pricing page adds attack surface management, dark web monitoring and weekly automated vulnerability scanning to two comprehensive penetration tests a year. If you want to know which shadow assets an attacker could reach today, talk to us about continuous testing.

About The Author

Mohammed Khalil is a Cybersecurity Architect at DeepStrike, specializing in advanced penetration testing and offensive security operations. With certifications including CISSP, OSCP, and OSWE, he has led numerous red team engagements, focusing on cloud security, application vulnerabilities, and adversary emulation. His work involves dissecting complex attack chains and developing resilient defense strategies for clients in the finance, healthcare, and technology sectors.

Sources and References

• IBM and Ponemon Institute, Cost of a Data Breach Report 2026 (29 July 2026): https://www.ibm.com/reports/data-breach
• IBM, Cost of a Data Breach Report 2026, full report PDF: https://assets.ibm.com/adobe/assets/urn:aaid:aem:21111142-1251-4369-86fb-57b82f5bb108/original/as/Cost%20of%20a%20Data%20Breach%20Report%202026.pdf
• IBM Newsroom, "IBM Study: One in Four Malicious Breaches are AI-Enabled" (29 July 2026): https://newsroom.ibm.com/2026-07-29-ibm-study-one-in-four-malicious-breaches-are-ai-enabled,-costing-companies-6-million-on-average
• IBM, Cost of a Data Breach Report 2025: https://www.ibm.com/downloads/documents/us-en/131cf87b20b31c91
• IBM Newsroom, Cost of a Data Breach Report 2024 release (30 July 2024): https://newsroom.ibm.com/2024-07-30-ibm-report-escalating-data-breach-disruption-pushes-costs-to-new-highs
• Verizon, 2026 Data Breach Investigations Report, executive summary: https://www.verizon.com/business/resources/executivebriefs/2026-dbir-executive-summary.pdf
• Verizon, 2025 Data Breach Investigations Report, executive summary: https://www.verizon.com/business/resources/reports/2025-dbir-executive-summary.pdf
• Gartner, "Gartner Unveils Top Eight Cybersecurity Predictions for 2023-2024" (28 March 2023): https://www.gartner.com/en/newsroom/press-releases/2023-03-28-gartner-unveils-top-8-cybersecurity-predictions-for-2023-2024
• Gartner, "Gartner Identifies Critical GenAI Blind Spots That CIOs Must Urgently Address" (19 November 2025): https://www.gartner.com/en/newsroom/press-releases/2025-11-19-gartner-identifies-critical-genai-blind-spots-that-cios-must-urgently-address0
• Microsoft and LinkedIn, 2024 Work Trend Index, "AI at Work Is Here. Now Comes the Hard Part" (8 May 2024): https://www.microsoft.com/en-us/worklab/work-trend-index/ai-at-work-is-here-now-comes-the-hard-part
• Cisco, 2025 Cybersecurity Readiness Index (7 May 2025): https://newsroom.cisco.com/c/r/newsroom/en/us/a/y2025/m05/cisco-study-reveals-alarming-deficiencies-in-security-readiness.html
• Netskope Threat Labs, Cloud and Threat Report 2026 (6 January 2026): https://www.netskope.com/resources/cloud-and-threat-reports/cloud-and-threat-report-2026
• Zylo, 2026 SaaS Management Index announcement (29 January 2026): https://zylo.com/news/2026-saas-management-index
• CISA, Binding Operational Directive 23-01 (3 October 2022): https://www.cisa.gov/news-events/directives/bod-23-01-improving-asset-visibility-and-vulnerability-detection-federal-networks