According to new research from Cyberhaven Labs, drawn from real-world data movement events and active user counts across Cyberhaven's enterprise customer base between December 2025 and June 2026, Claude Desktop usage grew 1,233% between January and June 2026.
Over the same period, employee adoption of generative AI (GenAI) software as a service (SaaS) applications barely moved: after reaching 34.2% in December 2025, adoption held in a narrow 35.1% to 35.9% band from January through May, before finally rising to 37.7% in June, the first meaningful shift in six months. That plateau wasn't uniform either. Employees at companies with 100 to 1,500 employees used GenAI SaaS applications at a substantially higher rate (42.3% in June) than employees at companies with more than 1,500 employees (31.5%).
These trends show that the tools security teams have spent a year learning to monitor and govern are consolidating, with security and productivity consequences. The tools running directly on employee devices, with direct access to local files and the ability to execute actions, are the ones now moving the fastest, and in some cases, the least visible ones.
What Is Agentic AI Adoption?
Agentic AI adoption refers to employee use of AI applications that run directly on endpoint devices and can take action on a user's behalf, rather than operating through a browser-based chat interface. This category includes desktop applications like Claude Desktop and Microsoft Copilot, along with AI coding assistants. Unlike browser-based GenAI SaaS tools, these applications can read local files, execute commands, and move data with a level of system access that chat interfaces do not have.
This tech adoption has accelerated sharply: endpoint agentic AI adoption roughly doubled year over year, reaching 60.5% among developers at its May 2026 peak. That growth is giving rise to the agentic enterprise, where human-to-human workflows are increasingly replaced by human-to-agent and agent-to-agent workflows.
Why Is Claude Desktop Usage Growing Faster Than ChatGPT Desktop or Microsoft Copilot?
Looking at Claude Desktop, ChatGPT Desktop, and Microsoft Copilot Desktop as a category, active users grew 43.2% between January and June 2026. That growth was not evenly split. Claude Desktop usage grew 1,233%, while ChatGPT Desktop declined 33.2% and Microsoft Copilot Desktop declined 81.8%.
Part of that gain reflects a small January base: Claude Desktop started the year with far fewer active users than its competitors, so a large percentage increase was easier to produce. But the direction of the shift still matters. Autonomous, agentic applications, not chat interfaces, are becoming the default way employees interact with AI on the endpoint. Each of these applications can read local files, execute actions, and move data with a level of system access that browser-based GenAI SaaS tools do not have. A security team that tracked only chat tool adoption in January would have missed where the real exposure was building by June.
That same pattern, a smaller set of tools carrying a growing share of activity and risk, shows up on the SaaS side too, just measured differently.
Data Movement Grew 80% Through GenAI SaaS
The number of distinct GenAI SaaS applications used by the average enterprise fell from 50.5 in the first half of 2025 to 47.6 in the first half of 2026. At the same time, data movement events into and out of GenAI SaaS, uploads, downloads, and copy-paste actions, grew 80% year over year between June 2025 and June 2026.
Fewer applications are now carrying far more traffic. Enterprises appear to be consolidating around a smaller set of sanctioned tools rather than reducing AI usage overall. The result is concentration risk: a smaller number of applications, each handling a larger share of an organization's data movement, whether that application is a browser-based GenAI SaaS tool or an endpoint agent like Claude Desktop.
Adoption Rate Alone Doesn't Measure AI Risk
Adoption percentage and usage intensity are not the same signal. A tool with a smaller or slower-growing share of employees can still generate the majority of an organization's agentic activity, and therefore carry the majority of its data exposure. Two organizations with identical GenAI SaaS or agentic AI adoption rates can carry meaningfully different risk profiles, depending on which applications employees actually run, how often, and with what level of system access.
This is the gap most AI governance programs still have. A program built to track which GenAI SaaS applications employees use, and at what rate, needs a second line of sight into which agentic applications run directly on the endpoint and with what system access, not just whether they are installed.
How Cyberhaven Addresses Agentic AI Risk
Cyberhaven is the leader in data security for the agentic enterprise. Cyberhaven traces the full lifecycle of data, adapting protection as context changes, connecting lineage, identity, and behavior to protect data in the context of how it is created, copied, fragmented, and shared across workflows, including the workflows agentic tools like Claude Desktop and coding assistants now touch. Operating in real time where humans and agents act on data, Cyberhaven helps stop risk before it becomes loss, giving security teams a way to act on system-level exposure that adoption metrics alone cannot show.
The Shift Ahead
Six months of data show two versions of the same story: GenAI SaaS consolidating into fewer, busier tools, and endpoint agentic AI concentrating activity into applications like Claude Desktop that most governance programs were not built to watch. Both point to the same blind spot heading into the second half of 2026: adoption metrics without workflow-level visibility.
Explore this data in-depth with the 2026 AI Adoption & Risk Report: Mid-Year Update.
Put this knowledge into action with “Securing AI Systems: An Enterprise Defense Framework.”
Frequently Asked Questions
How much did Claude Desktop usage grow in 2026?
Claude Desktop active users grew 1,233% between January and June 2026, according to Cyberhaven Labs data drawn from its enterprise customer base.
Why did Claude Desktop usage grow while ChatGPT Desktop and Copilot declined?
Claude Desktop's growth started from a small January base, which made a large percentage increase easier to produce. ChatGPT Desktop declined 33.2% and Microsoft Copilot Desktop declined 81.8% over the same period.
How much has data movement through GenAI SaaS grown year over year?
Data movement events into and out of GenAI SaaS, including uploads, downloads, and copy-paste actions, grew 80% between June 2025 and June 2026, even as the number of distinct applications per enterprise declined.
Why is endpoint agentic AI riskier than browser-based GenAI SaaS?
Endpoint agentic applications can read local files, execute actions, and move data with system-level access that browser-based GenAI SaaS tools do not have, which widens the potential exposure of any single application.
How does data lineage help secure agentic AI workflows?
Data lineage traces how sensitive data is created, copied, and shared across applications, endpoints, and agents, giving security teams visibility into system-level activity that adoption-rate metrics alone cannot capture.
.avif)
.avif)
