Fastly is expanding its edge platform with three AI-focused security and governance capabilities designed to give enterprises — and the MSPs and MSSPs supporting them — more control over how AI models, applications and autonomous agents interact with enterprise infrastructure.
The new AI Runtime Control, AI Firewall and API Security capabilities are intended to apply routing, security, access and governance policies as AI requests occur, rather than relying primarily on after-the-fact monitoring. Fastly says the need is growing quickly: machine-generated traffic accounted for more than half of traffic across its network in July and August, while AI traffic grew 6.5 times faster than human traffic from January through May.
For channel partners, the shift creates a potential new managed-services layer around AI visibility and governance. As customers adopt multiple models and autonomous agents, MSPs and MSSPs may increasingly be asked to track which AI systems are being used, control what agents can access, and help manage the security and cost implications of machine-speed activity.
Fastly brings AI controls into the request path
Fastly’s new capabilities extend the company’s existing edge platform to AI workloads. The tools are designed to apply security, access, routing, and governance policies as requests happen rather than analyzing activity only after the fact.
The company is targeting three paths in particular: applications calling AI models, users and systems interacting with AI applications, and AI agents accessing enterprise APIs.
Speaking with Channel Insider, Fastly VP of Partnerships Jeff Alpen said the underlying problem was less about enterprises lacking individual security or management features and more about responsibility for AI activity becoming fragmented across different teams.
“The gap wasn’t a missing feature so much as a missing owner for the layer,” Alpen said. “Finance couldn’t explain why AI spend jumped in a quarter, engineering teams each built their own failover, and none of them worked the same way, and security teams had no way to inspect a prompt before it reached a model.”
Fastly adds model-agnostic AI controls
According to Alpen, many existing tools also sit outside the request path, meaning they can report on activity after it occurs but have less ability to govern requests in real time.
Fastly is extending its edge platform into that control layer without requiring customers to standardize on a particular AI provider. The company does not host or operate the underlying models, allowing organizations to retain their existing provider relationships and infrastructure choices.
The three capabilities announced by Fastly are:
- AI Runtime Control: Routes model calls through a single endpoint across public and self-hosted providers. It includes virtual keys designed to protect provider credentials, as well as visibility into token spend, rate limiting, budget controls, and failover.
- AI Firewall: Evaluates prompts in the request path to identify and block LLM-focused attacks, including prompt injection, before malicious requests reach the targeted model.
- API Security: Enforces API contracts across agentic, agent-assisted, and conventional traffic. Organizations can observe or block requests that do not conform to established API rules on a service-by-service basis.
All three capabilities are now available, according to Fastly.
AI agents challenge conventional API security assumptions
The challenge becomes more complicated when AI agents begin interacting directly with enterprise APIs. Unlike conventional human or application traffic, agents can aggressively retry requests, operate across multiple services in parallel, and take actions at machine speed.
Alpen said those characteristics create new concerns about access, cost, reliability, and governance. They also complicate the assumptions underlying traditional API monitoring.
“Agent traffic breaks most of the assumptions that conventional API monitoring is built on because they don’t behave like conventional human or application traffic and are harder to predict and audit,” Alpen said.
He pointed to aggressive retries, parallel requests across multiple services, and sequences of API calls that would be unusual for a conventional human session but may be normal behavior for an autonomous agent.
Alpen argues that authorization presents another complication.
“They also sit awkwardly in existing authorization models, because an agent acting on a user’s behalf is neither cleanly a user nor cleanly an application, and most systems assume it’s one or the other,” Alpen said.
That distinction becomes increasingly important when agents can perform actions rather than simply retrieve information. Alpen also said that after-the-fact review becomes less useful once those calls have already been made.
Fastly’s API Security capabilities are designed to enforce existing API contracts as agent requests occur, allowing organizations to observe or block non-conforming requests across individual services.
AI growth creates cost and governance pressure
Security is only one side of the control problem Fastly is targeting. As enterprises move AI projects from pilots into production, different development teams may use different models and providers, creating challenges around cost visibility, routing, failover, and policy enforcement.
That fragmentation can make it difficult for finance, engineering, and security teams to maintain a consistent view of AI usage across the organization.
“As AI traffic grows, organizations need real-time control over which AI systems are used, how much they cost, what AI agents are allowed to do, the quality of their responses, and how reliably services perform as demand grows,” Alpen said.
AI Runtime Control is intended to centralize some of those functions while keeping the underlying models separate. Model calls can be routed through a common endpoint, where organizations can apply rate limits, budget controls, routing policies, and failover across public or self-hosted providers.
The approach reflects a broader multivendor reality in enterprise AI. Organizations may continue to use multiple models and providers while seeking a common layer to apply governance and operational controls across them.
Fastly sees a managed AI governance opportunity for MSPs
For MSPs and MSSPs, the expansion of enterprise AI introduces systems and activity that may sit outside the infrastructure they traditionally deploy and manage.
“MSPs and MSSPs already know what’s deployed, exposed, and under attack in their customers’ environments, but AI adds a tier they didn’t provision and can’t see, with teams calling models directly on credentials and budgets nobody is tracking,” Alpen said.
That could expand the scope of managed services beyond monitoring infrastructure and applications.
According to Alpen, partners may increasingly need visibility into which models customers are accessing, who is calling them, and how much those interactions cost. Governance could also include determining which providers are approved and what actions AI agents are permitted to perform.
Security responsibilities similarly extend beyond the traditional application perimeter. Prompts sent to models and API requests initiated by agents can become part of the environment MSPs and MSSPs are asked to monitor.
“The unit of management shifts from infrastructure to behavior, and because agents act at machine speed, those controls have to be enforced at runtime, in the request path,” Alpen said.
Governance without ownership
For service providers, that creates a possible managed governance role without requiring them to take ownership of customers’ underlying AI models. Customers could maintain their relationships with model providers while an MSP or MSSP operates the control layer across them.
That model could give service providers another managed security and governance function as customers put more AI workloads into production. Whether it develops into a distinct managed service will depend in part on how broadly enterprises adopt autonomous agents and how much AI governance they ultimately hand to outside providers.
KDDI America and KDDI Europe are adding Exaforce’s agentic SOC platform to their managed security portfolios across the US and EMEA. Read more about how the partnership expands Exaforce’s reach into multinational enterprises and managed security environments.