Cloudflare spotlights agentic AI infrastructure and security at LEAP 2026

Cloudflare used LEAP 2026 in Riyadh to lay out a practical, security-first approach for managing AI workloads and the new class of agentic infrastructure emerging in enterprise environments. The company presented its roadmap for protecting systems where autonomous agents act on behalf of organizations, and where those agents’ behaviors and interactions require the same identity and permission controls historically applied to human users. The demonstration at LEAP, running August 31 to September 3, framed the challenge as distinct from securing passive machine-learning models: active agents execute actions, make requests, and interact with other systems, and therefore expand the attack surface in novel ways. Central to Cloudflare’s message was the need to treat autonomous agents as first-class identities within corporate security models. The company highlighted Cloudflare One as a cornerstone for applying Zero Trust principles to agentic infrastructure, ensuring continuous authentication and authorization for every interaction an agent initiates. By extending the same rigor used for employees and contractors to software agents, organizations can enforce least-privilege access, log decisions, and revoke capabilities as circumstances change. That approach reframes agents not as ephemeral processes but as entities requiring governance, auditability, and lifecycle management. Cloudflare also underscored tooling designed to prevent some of the most specific risks that target models and their interfaces. The company described Firewall for AI as a purpose-built layer to defend against attacks such as prompt injection—malicious inputs crafted to alter an agent’s behavior or expose sensitive data. Complementing the firewall, AI Gateway was presented as a traffic-control plane for AI workloads, enabling organizations to inspect, filter, and route agent-originated requests between services and external endpoints. Together with Cloudflare Workers, which provides a serverless execution environment at the network edge, these components allow security policies and custom logic to be enforced with low latency, close to where agents operate. The focus on agent identity and permissioning is not merely theoretical; it responds to an observable shift in how organizations are deploying AI. Ercan Aydin, Cloudflare’s Area Vice President for the Middle East, Turkey & Africa, emphasized that in Saudi Arabia the technology is moving rapidly from experimentation into production. As organizations transition to operational deployments, the expectations for uptime, compliance, and risk management increase. That makes governance features and transparent telemetry essential, because production-grade agentic systems must meet the same standards for accountability and regulatory compliance as other critical infrastructure. Cloudflare’s presentations at LEAP made clear that operational AI introduces both functional and security complexity. Agents that autonomously browse, query, and act can inadvertently leak secrets, execute unintended transactions, or be manipulated into carrying out harmful tasks. Treating those agents as identities with defined permissions mitigates these risks by constraining the scope of their actions, applying robust authentication, and making their behavior observable through centralized logs and policy controls. This identity-centric posture also supports incident response: when an agent behaves anomalously, administrators can trace its activity, isolate its access, and remediate configurations without having to take down broader services. Another strand of Cloudflare’s LEAP coverage addressed how content providers and publishers can exert control over how AI systems consume and use their material. The company introduced AI Crawl Control and Pay Per Crawl as mechanisms to regulate access to web content by AI consumers. These features aim to give publishers fine-grained options to allow, throttle, or monetize automated content ingestion so that data owners retain agency over how models see and repurpose their material. In the context of agentic systems that may scrape or query large swaths of online content autonomously, these controls offer publishers a way to assert policy and economic choices over AI interactions. Cloudflare tied its product messaging to hard trends in the threat landscape. The company’s H1 2026 DDoS Threat Report revealed a surge in distributed denial-of-service attacks, with some incidents reaching terabit-per-second scale. Such volumetric attacks and increasingly complex, multi-vector campaigns underscore the need for automated, scalable defenses that can respond at network speed. Cloudflare argued that defending agentic environments requires similar automation and scale: both to protect agents from being abused as attack vectors, and to shield services that agents depend on from disruption. The implication is that operators should deploy defenses that are capable of contextual decision-maki