Fastly launches AI Firewall and AI Runtime Control to secure AI traffic at the network edge

Fastly launches AI Firewall and AI Runtime Control to secure AI traffic at the network edge

The edge company is expanding its Fastly for AI portfolio with AI Runtime Control, AI Firewall and new API security, intended to give real-time control over model calls, prompt injection and agent traffic. The launch came just before an investor day that set a revenue target of $1.1–1.3 billion in 2029 — and sent the stock sharply up and down across two days.

What happened

On Monday, September 21, 2026, Fastly launched two new capabilities, AI Firewall and AI Runtime Control, intended to give customers better protection and monitoring of their AI systems, according to The Motley Fool. The same day, the company — by its own account, relayed through Help Net Security — expanded the Fastly for AI portfolio with three components: AI Runtime Control, AI Firewall and new API Security features.

The next day, Tuesday, September 22, the company held its investor day at the Nasdaq MarketSite in New York, where management presented platform strategy, product roadmap and financial targets. The revenue target was set at $1.1–1.3 billion in 2029, according to CoinCentral and 24/7 Wall St.

The timing is no accident. The launch gives the market narrative concrete products to hang on it: Fastly is positioning itself as the control layer for AI traffic at the edge of the network, at the same time as the company asks investors to believe in a substantial growth story through 2029.

How the products are meant to work

All product descriptions below are Fastly's own claims, relayed through Help Net Security's coverage of the company's announcement — not independently verified facts.

AI Runtime Control is intended to govern AI access and usage. Model calls are routed through a single endpoint across public and self-hosted model providers. The company highlights so-called virtual keys that are meant to protect the underlying provider credentials. According to Fastly, the capability provides visibility into token spend, rate limiting, budget management and failover between models. Concretely, this means a company can switch between commercial and internal models without changing its application code, while costs can be tracked and capped centrally.

AI Firewall is intended to protect AI applications against LLM-based attacks, including prompt injection, by evaluating prompts directly in the request path at the edge of the network. That places the control in real time within the traffic flow itself, rather than after the fact or in a separate inspection service.

API Security is intended to enforce API contracts when AI agents access enterprise APIs. The rules are meant to apply consistently across three traffic types: agent-driven traffic, agent-assisted traffic and conventional traffic. According to Fastly, organizations can choose to observe or block anomalous requests, service by service.

It is worth noting what has not been disclosed: Fastly has, according to the available coverage, not published pricing, availability or named customers for the new products.

The numbers behind the positioning

Fastly itself highlighted two network measurements as the framing for the launch: machine-generated traffic crossed 50 percent of traffic on Fastly's network in July and August this year, and AI traffic grew 6.5 times faster than human-generated traffic from January to May. Both figures are Fastly's own measurements and have not been independently confirmed.

The company also cites McKinsey for the claim that 93 percent of organizations exceed their AI budgets. This figure, too, is known only through Fastly's announcement; the underlying McKinsey source is not present in the available source material.

The company's fundamental numbers do at least support that the security business is growing: in the second quarter, revenue rose 23 percent to $183 million, and remaining performance obligations (RPO) rose 38 percent to $341 million, according to Invezz. Security revenue grew 43 percent year over year, according to 24/7 Wall St.

The market: two volatile days, and split analyst conviction

The stock reaction was as dramatic as it was divided. According to CoinCentral, the stock rose 14 percent on Wednesday, closing at $29.65 after the investor day. The day before, on Tuesday, the stock had fallen 7 percent, while competitors Cloudflare and Akamai rose slightly — a decline 24/7 Wall St. described as profit-taking after a steep run-up beforehand.

The analysts landed in different places. RBC Capital maintained a Sector Perform rating and a $25 price target, pointing, according to CoinCentral, to Tuesday's decline as a sign that investors were not fully convinced by the wide 2029 revenue range. Evercore ISI kept Outperform with a $32 price target, while BofA kept Underperform but raised its price target to $22 from $20. The spread from $22 to $32 reflects genuine disagreement over whether the AI positioning and the 2029 targets hold up.

At the investor day, CEO Kip Compton described, according to 24/7 Wall St., the edge as a complementary layer to hyperscale cloud infrastructure, not a replacement for it. It is a deliberate distance from the largest cloud platforms — Fastly wants to sit between users and models, not compete to train or run them.

What is verified, and what is not?

This is a product launch covered through secondary sources, and a distinction matters between what is established and what is claimed:

• Established: The launch date (September 21), the expansion of the Fastly for AI portfolio with the three named components, the September 22 investor day with 2029 targets of $1.1–1.3 billion, and the Q2 figures for revenue, RPO and security growth.
• Fastly's own claims: All product capabilities — endpoint routing, virtual keys, budget management, request-path evaluation of prompts, per-service API contract enforcement.
• Fastly's own measurements: The figures on 50 percent machine traffic and 6.5-times-faster AI growth.
• Unresolved: The AI Firewall's effectiveness against prompt injection has not been tested or verified by any independent party in the available material. Pricing and availability have not been disclosed. The stock price picture varies across sources, reflecting different snapshots of different trading days, and the figures should be read as separate data points — not averaged.

A claim is also circulating that Fastly reportedly handles a significant share of Meta's Muse agent traffic. This is unverified secondhand coverage and should not be treated as confirmed.

What would validate the positioning

Fastly has placed three claims on the table: that machine traffic already dominates its network, that enterprises need control in the request path, and that the edge is the right place to put that control. For this to land with both customers and the market, three things are needed that do not yet exist: named customers actually putting AI Runtime Control and AI Firewall into production, disclosed pricing and availability, and independent security testing — particularly of the claim of real-time prompt injection mitigation, which is the most technically demanding and commercially valued capability. Until then, this is a compelling narrative with a well-timed launch, but with effectiveness and willingness to pay still unproven.