Enterprise AI Adoption Statistics: 509% Growth in One Year

Security programs tend to scope AI monitoring to the tools employees use most. Cyberhaven Labs data suggests that's the wrong filter. Between February 2025 and February 2026, endpoint AI app usage grew 509% across an enterprise population of hundreds of thousands of employees, and the fastest growth sits in tools with direct access to source code, credentials, and customer data. Popularity tells you where usage is today. Growth rate and data access tell you where exposure is heading, and governance programs built for last year's footprint now cover a much larger one.

How Fast Is Enterprise AI Adoption Growing?

Endpoint AI app usage grew 509% between February 2025 and February 2026, roughly sixfold, according to Cyberhaven Labs data from hundreds of thousands of employees. Coding assistants grew 357%, and Claude grew 5,680%. The categories growing fastest also have the deepest access to source code, credentials, and customer data.

The three figures measure different things. The 509% figure is the aggregate across all endpoint AI native apps, 357% is the coding assistant category, and 5,680% is a single tool. Read together, they show growth across the whole surface, with the largest increases concentrated in tools that have privileged access to enterprise data.

A sixfold expansion in 12 months changes the scale of what security teams have to govern. A program sized for February 2025 now covers roughly six times as many users of endpoint AI apps.

Methodology: Endpoint AI native apps are AI applications such as Claude, ChatGPT, and Copilot that run as desktop apps on an employee's device. Growth figures reflect year-over-year change in enterprise users in the Cyberhaven Labs dataset, which covers hundreds of thousands of employees.

Why Are Coding Assistants the Highest-Risk AI Category?

Coding assistants are AI applications that read, generate, and modify enterprise code in real time, usually with direct access to a developer's local environment, version control system, and internal repositories. Adoption grew 357% year over year, the fastest of any category in the Labs dataset.

Risk follows access. Coding assistants read files, traverse directory structures, and, in agentic configurations, execute commands. The data within reach includes:

• Source code containing proprietary logic and product intellectual property (IP)
• Credentials and API keys in configuration files and environment variables
• Internal architecture details referenced in comments and linked files
• Build and deployment scripts that describe production infrastructure

Generative and agentic use create different exposure. In generative use, a developer pastes code into a prompt, and the exposure is the code shared. In agentic use, the tool browses a local repository without a prompt for each file, so exposure depends on what the tool can read. General AI policy doesn't account for either pattern, and source code, credentials, and API keys need targeted detection rules.

Security teams without detection and policy coverage specific to coding assistants have a blind spot in the fastest-growing category.

Why Do Employees Use So Many Different AI Models?

Employees pick the model that fits the task, and they don't wait for IT to sanction it. The Labs data shows no convergence on a single tool. Claude's usage grew 5,680% year over year, and growth like that tends to follow a model proving useful for specific work, often ahead of any formal approval process.

The operational consequence is a visibility gap. A monitoring strategy built around ChatGPT misses Claude activity entirely. A Copilot-focused policy doesn't cover an engineer on a Mac who opens the Claude desktop app to debug production code. Each unmonitored tool removes data that security teams need to detect exfiltration, enforce policy, and investigate incidents.

Model diversity also shortens the shelf life of any approved-tool list. A tool absent from last quarter's inventory can hold meaningful usage this quarter, so coverage defined by a fixed list ages quickly. Visibility across all models, both generative and agentic, is now a baseline requirement for any AI security program. For more on this pattern, see the analysis of fragmented AI adoption and data risk.

Where Does Enterprise AI Usage Concentrate?

Usage concentrates by operating system and by department, and both patterns affect where detection effort should go.

Mac vs. Windows Usage

Microsoft Copilot accounts for 54% of endpoint AI usage overall in 2026, but that share hides a sharp platform split.

Most enterprises run mixed fleets, so the tools in use, the data access they create, and the resulting risk profile differ by operating system. A governance approach built around Microsoft's native tooling covers one platform segment well and misses most endpoint AI activity on Mac.

Engineering and Sales Drive Most AI Usage

Engineering accounts for 33% of AI native app usage across the enterprises in the dataset, and sales accounts for 25%. Together, the two functions drive 58% of total usage.

Both functions hold data that is costly to lose. Engineering holds source code, infrastructure credentials, and product IP. Sales handles customer data, contracts, pricing, and deal information. Coding assistants, the fastest-growing category, are disproportionately an engineering tool. The practical response is to calibrate detection and response to where the sensitive data sits while keeping AI available to both teams.

Labs data on financial services, manufacturing, and professional services shows how usage varies by industry.

What Should Security Teams Do About Fast-Growing AI Tools?

The Labs data points to three operational gaps worth testing against your current program.

1. Coverage across tools. Monitoring scoped to a single vendor or model produces an incomplete picture. Claude's 5,680% growth shows how quickly that picture changes, so coverage should extend to every AI tool, both generative and agentic, rather than the most popular one.
2. Coding assistant detection. The access patterns coding assistants create need their own rules. Source code, credentials, and API keys require targeted detection that covers interactive use and autonomous repository access.
3. Endpoint visibility across operating systems. Cloud and SaaS-layer monitoring misses native app activity on the endpoint, along with the Mac and Windows differences in which tools dominate.

A practical first step is to compare the AI tools your program covers against the tools employees run on their devices. Any tool in the second list and missing from the first is an unmonitored path for source code, customer data, and credentials.

The Governing the Autonomous Enterprise framework covers how to structure that coverage for agentic AI, and the AI governance guide covers the policy side.

How Cyberhaven Addresses AI Adoption Risk

Cyberhaven approaches AI adoption risk as part of Data Security for the Agentic Enterprise. AI Security gives security teams visibility into which AI tools, both generative and agentic, are in use across the organization, including native apps on the endpoint. Data Lineage records where sensitive data originated and how it moved, so a source code file or credential reaching a coding assistant is traced to its origin and judged on its history, not on content alone.

This coverage applies across Mac and Windows, which matters because the AI tools employees favor differ by platform. The same lineage record supports detection, policy enforcement, and investigation, so one view of data movement serves all three.

Endpoint AI adoption grew roughly sixfold in a year, and the fastest growth sits in tools that read source code, credentials, and customer data. Security teams that extend visibility across every tool, model, and operating system can close that gap while policies are still being written. Cyberhaven traces sensitive data into generative and agentic AI tools on the endpoint, so detection keeps pace with adoption

Learn more about AI adoption trends with The 2026 AI Adoption & Risk Report: Mid-Year Update.

Frequently Asked Questions

How fast is enterprise AI adoption growing?

Endpoint AI app usage grew 509% between February 2025 and February 2026, according to Cyberhaven Labs data from hundreds of thousands of employees. That's roughly a sixfold increase in 12 months. Coding assistants grew 357% over the same period, and Claude grew 5,680%.

Why are coding assistants considered the highest-risk AI category?

Coding assistants read files, traverse directory structures, and in agentic configurations execute commands. The data they can reach includes source code, credentials, API keys, internal architecture details, and deployment scripts. That combination of privilege and autonomy gives ungoverned access in this category the highest potential for harm.

What does endpoint AI adoption mean, and why does it matter for security?

Endpoint AI adoption refers to employees using AI native apps, such as Claude, ChatGPT, and Copilot, as desktop applications on their devices rather than through a browser. These apps can reach local files, environment variables, and system resources that cloud-based tools can't. Security teams that monitor only SaaS-layer activity miss this entire class of usage.

How can security teams see which AI tools employees are using?

Visibility requires monitoring at the endpoint across all AI tools and models, including the ones IT hasn't approved. Data lineage tracks data from its source to its destination, including AI tools, which lets security teams detect exfiltration, enforce policy, and investigate incidents across a mixed AI footprint.

Why does operating system fragmentation affect AI security strategy?

Different AI tools dominate on different platforms. ChatGPT is 40 times more popular than Copilot on Mac, while Copilot is 7.6 times more popular than ChatGPT on Windows. Enterprises with mixed fleets carry different AI risk profiles across their endpoints, so governance built around one vendor leaves gaps.

Which AI tools are enterprises adopting fastest in 2026?

Coding assistants are the fastest-growing AI category in the Cyberhaven Labs dataset, at 357% year-over-year growth. Among individual tools, Claude grew 5,680%. Microsoft Copilot holds the largest share of endp