Cloudflare says traffic generated by artificial intelligence agents has climbed more than 1,700% in the past year and now makes up about 60% of total traffic on its network. The company shared the figures during a media briefing in Seoul, pointing to a major change in how internet activity is being created and carried out.

Goran Risticsevich, Cloudflare’s executive vice president and managing director for Asia-Pacific, said the internet was originally built for people. Now, AI agents are producing traffic at machine speed.

AI Agent Traffic Is Reshaping Cloudflare’s Network

The reported growth shows how quickly AI agents are taking on tasks across online services. Rather than traffic coming only from people browsing websites and using applications directly, more requests now come from automated agents working on behalf of users.

Cloudflare said this AI-generated activity has increased by more than 1,700% over the past year. At roughly 60% of the company’s total traffic, AI agents now account for a substantial share of activity on its network.

That shift also changes the security picture. AI agents can access multiple services at the same time, which expands the number of systems and connections that organizations need to manage.

Cloudflare Warns About Shadow AI and Shadow MCP

Cloudflare identified “shadow AI” as a growing concern. The term refers to employees using AI tools for work without approval, which can expose sensitive internal data to external models.

The company also warned about risks tied to rapid adoption of the Model Context Protocol, or MCP. MCP connects AI applications with external tools and internal systems. But unauthorized protocol connections can create “shadow MCP” risks by bypassing corporate controls.

These issues are linked to the wider use of AI tools inside organizations. When tools, prompts, or protocol connections are used outside approved processes, internal data and systems may be exposed in ways that organizations have not accounted for.

Cloudflare recommended several steps for organizations managing AI use internally:

  • Identify the AI tools being used within the organization.
  • Block prompts containing sensitive information before they reach AI models.
  • Apply zero-trust policies to MCP server access.
  • Limit access so users can reach only the applications required for their roles.

The guidance focuses on visibility and access control. Organizations first need to know which AI tools and connections are in use, then control what information can be sent and what systems each user can access.

Post-Quantum Cryptography and “Q Day”

Cloudflare also discussed post-quantum cryptography. Anika Gabers, head of Cloudflare One go-to-market, said the company expects “Q Day” — the point at which quantum computers can break current encryption — to arrive around 2030.

Cloudflare already supports post-quantum cryptography and plans to provide full quantum-resistant authentication by 2028.

The company’s comments place AI agent growth and encryption planning alongside one another. As automated systems create more network activity and interact with connected tools, access controls and authentication remain part of the broader security discussion.

Cloudflare Kitesurf Is Built for AI Agents

Cloudflare launched Kitesurf on August 6. It is a lightweight browser built from scratch for AI agents rather than human users.

Unlike traditional browsers, Kitesurf removes visual features such as tabs and extensions. It runs entirely inside V8 isolates on Cloudflare’s Workers platform.

Cloudflare says Kitesurf uses far less CPU and memory than Chromium for common agent tasks, including screenshots and HTML extraction. The browser is available in a free beta.

Kitesurf arrives as Cloudflare reports rapid growth in AI agent traffic. One announcement focuses on infrastructure built specifically for agents, while the other focuses on the access controls and security policies needed as those agents interact with services and systems.