Fastly launches AI firewall & runtime control tools
Tue, 22nd Sep 2026 (Today)
Fastly has launched AI Runtime Control, AI Firewall and new API Security features, expanding its Fastly for AI product set.
The new tools are designed to give organisations real-time visibility and control across AI systems as businesses move AI use from trials into production. They focus on three areas: model access and usage, protection for AI applications, and rules governing how AI agents connect to enterprise APIs.
Fastly is framing the launch around a sharp rise in machine-driven traffic on its network. According to the company, machine-generated traffic rose above 50% in July and August, while AI traffic grew 6.5 times faster than human traffic from January through May.
Fastly said that increase is creating operational and cost pressures for companies deploying AI services. Citing McKinsey research, it said 93% of organisations are exceeding their AI budgets as they try to manage model routing, spending, authorisation and reliability.
Three controls
AI Runtime Control is designed to sit between applications and AI models. It routes model calls through a single endpoint across both public and self-hosted providers, while using virtual keys to shield provider credentials.
The feature also includes visibility into token spending, rate limiting, budget controls and failover. This allows organisations to use different models and infrastructure providers while applying central policy controls in the request path.
AI Firewall focuses on AI applications themselves. It is designed to mitigate attacks on large language model systems, including prompt injection attempts, by evaluating prompts directly in the request path before they reach the model.
The new API Security functions are intended to address the growing role of AI agents and AI-assisted software development. Fastly said organisations are creating new endpoints faster than they can be documented or reviewed, raising the risk that agents could call unsupported operations or submit malformed requests.
Under the new API Security offering, customers can enforce API contracts and apply the same rules across agentic, agent-assisted and conventional traffic. The system can either observe or block requests that do not conform to policy on a service-by-service basis.
Unified platform
The launch broadens Fastly's effort to position AI governance and application security on the same platform it already uses for content delivery, edge compute and security services. Existing tools in that wider portfolio include Bot Management, ContentGuard, DDoS Protection, API Discovery and Next-Gen WAF.
Fastly said customers can now apply consistent access, security and routing policies across vendors or model providers without giving up model choice. In practice, this reflects a growing concern among enterprises that AI systems will be assembled from multiple external and internal services rather than a single model supplier.
The company also used the announcement to underline the scale of its infrastructure, saying its network runs at 622 Tbps of global edge network capacity and handles more than five trillion requests a day.
Kelly Shortridge, Chief Product Officer at Fastly, said the shift to AI in production is changing what enterprise customers need from infrastructure and security suppliers.
"To effectively adopt coding agents and implement AI features without losing velocity or eroding existing business resilience, enterprises need control in production, at runtime, without delay, friction, or disruption," said Kelly Shortridge, Chief Product Officer at Fastly.
She said the company was extending controls already used for delivery and software security into AI systems and agents.
"Fastly is extending the real-time control customers already trust for content delivery and software security to the models, applications, and agents shaping this new era of distributed systems. Organisations deserve the freedom to move fast with AI while staying safe, achieving innovation at scale without trading off resilience," Shortridge said.
The products are now available.