Cloudflare bets big on secure AI agent infrastructure

Cloudflare is pioneering the 'Agent Cloud,' a novel computing foundation tailored specifically for AI agents rather than human users, addressing their unique demands for rapid processing, structured data access, and seamless system integration. By directly engaging developers and querying AI agents about their infrastructure needs—spanning storage, compute, execution, and secure data access—Cloudflare grounds its design decisions in authentic agent requirements, ensuring that the Agent Cloud is truly agent-native rather than a repurposed human-centric platform.

The Agent Cloud vision embodies a dual mandate: it must simultaneously serve as a foundational platform optimized for AI agents while bridging the existing human-centric web with the emerging agent-driven digital ecosystem. This duality reflects Cloudflare’s strategic approach to not only build the fundamental building blocks for an AI-native future but also to maintain interoperability with traditional tools, facilitating a smoother transition as autonomous agents evolve beyond simple chatbots into complex, integrated digital actors.

Positioning itself amid a burgeoning wave of agent-based AI innovation led by industry giants like OpenAI, Google, and Anthropic, Cloudflare’s Agent Cloud initiative stakes the company as a critical infrastructure enabler for the next generation of AI. By focusing on the underlying cloud services tailored for autonomous agents, Cloudflare is carving out a pivotal role in supporting the evolution of AI from conversational interfaces to sophisticated, autonomous agents that require specialized, high-performance infrastructure.

Cloudflare OS emerges as an open-source, browser-based AI workspace tailored for enterprises seeking secure, zero-trust environments that integrate AI agents with organizational data and workflows. By combining an agent workspace with a governance layer centered on Gatekeeper controls and personal app deployment via Cloudflare Workers, it allows organizations to operate entirely within their own systems and data, eliminating the need to expose broad API keys to AI agents. This design not only enhances security but also empowers enterprises to maintain strict oversight and fine-grained access control over AI-driven processes.

At the heart of Cloudflare OS’s security model is a zero-trust, capability-based framework where AI agents start with no permissions and must request access to specific resources mediated by Gatekeepers—specialized Cloudflare Workers that hold credentials and enforce scoped access. This approach ensures that credentials never touch the agents or their generated code, drastically reducing the risk of data leakage. Every interaction is logged and access is verified before users can view agent outputs or shared workspaces, addressing a critical enterprise concern about accidental exposure of sensitive data.

Cloudflare OS democratizes AI usage across enterprises by providing an intuitive, browser-based platform that supports both AI agents and human users, including non-engineers who might otherwise be excluded from developer-centric tools. Leveraging Cloudflare’s Dynamic Workers and Durable Object Facets, each AI-built app is a real Cloudflare Worker capable of automating tasks using tools created by users themselves, effectively turning AI into a versatile toolmaker. This design fosters productivity while embedding responsibility and governance, as users deploying agents remain accountable for their outputs, with managerial oversight ensuring continuity and control.

By packaging Cloudflare OS as a cohesive operating system for the AI-first enterprise, Cloudflare simplifies the integration of AI infrastructure within existing organizational ecosystems. The platform’s open-source architecture allows enterprises to customize AI models, branding, skills, and governance policies through Gatekeeper-controlled connectors, providing administrators with granular control over AI capabilities and human sign-off requirements. This infrastructure-first approach not only coexists with existing AI applications but also offers a clear, manageable path for enterprises grappling with the complexities of adopting AI at scale.

Cloudflare OS pioneers a zero-trust, capability-based security model that fundamentally redefines AI governance by never exposing raw credentials to AI agents. Instead, it employs Gatekeeper Workers—service-specific Cloudflare Workers—that mediate every access request, issuing scoped capability objects rather than broad API keys. This approach ensures that AI agents start with zero permissions and can only access explicitly authorized resources, effectively mitigating risks associated with shadow AI and uncontrolled AI spend, as broad, long-lived keys are replaced by dynamic, request-based access controls.

Comprehensive audit logging and strict access verification form the backbone of Cloudflare OS’s compliance and data safety strategy. Every AI agent action and resource read is meticulously logged, while user permissions are checked before opening workspaces or viewing agent outputs, preventing accidental data leaks—a common enterprise vulnerability. This rigorous oversight extends to assigning clear accountability: users and their managers are responsible for AI agent outputs, ensuring that governance is not only technical but also organizational, fostering a culture of responsibility and traceability.

Cloudflare OS balances innovation with enterprise compliance by embedding organizational context directly into AI workflows, emphasizing that understanding internal policies and data safety requirements matters more than the choice of AI model. Built atop Cloudflare’s Developer and Zero Trust platforms, the system integrates governed connectors and custom services to enforce least privilege access, ensuring users never have more permissions with AI than they do with direct system access. This design philosophy addresses enterprises’ regulatory concerns while enabling safe, scalable AI adoption.

Addressing cost control and vendor lock-in concerns, Cloudflare OS offers configurable budgets, rate limits, and delegated task management across AI models, enabling enterprises to tightly manage AI-related spending. Its open-source nature empowers organizations to inspect, customize, and integrate their own systems and AI models, preserving compliance and governance without surrendering control to external vendors. As Rita Kozlov emphasizes, this transparency is critical because 'you cannot put your company into software you do not own,' making Cloudflare OS a compelling foundation for AI-first enterprise infrastructure.

Cloudflare has strategically positioned itself at the nexus of security, networking, and developer tools to capture the burgeoning demand for secure, cloud-native AI infrastructure among enterprises. This unique intersection allows it to serve companies heavily experimenting with AI, differentiating itself from competitors by leveraging its core strengths in security and networking to address AI governance and observability needs. However, it faces stiff competition from security-centric firms like Zscaler and hyperscalers such as AWS and Azure, who are aggressively targeting the same governance and observability budgets within the AI infrastructure market.

The rollout of AI-centric products such as the AEO Visibility Dashboard, Cloudflare OS, and the Identity-Aware AI Gateway underscores Cloudflare’s commitment to evolving beyond traditional traffic management into deeper enterprise AI infrastructure roles. These innovations not only reinforce its narrative around AI-driven secure infrastructure but also clarify potential monetization pathways through usage-based and platform-style revenue models. Nevertheless, despite rising sales reported in Q2 2026, Cloudflare is grappling with widening net losses, highlighting ongoing challenges in balancing aggressive AI product expansion with cost discipline and profitability amid increased AI network usage.