# Huawei and Fastly launch products called an AI firewall, for entirely different jobs > Huawei and Fastly each launched a product called an AI firewall this month, and each protects something different. What each one does, and where a Mac fits in. FireAI Security & Research Team (HisnLabs) · Published 2026-09-28 Canonical: https://hisnlabs.com/en/news/what-is-an-ai-firewall-huawei-fastly-mac Two large companies launched something each calls an "AI firewall" within the same week, and the two products protect different things. Huawei published a technical white paper for a product named AIFW, [reported by Antara News](https://en.antaranews.com/news/433177/huawei-releases-an-aifw-technical-white-paper-to-facilitate-firewall-upgrade) on 27 September, and Fastly [introduced AI Runtime Control, including an AI Firewall feature](https://www.fastly.com/blog/introducing-fastlys-ai-runtime-control-why-your-ai-strategy-needs-control-plane), on 21 September. ## Background A firewall traditionally guards the edge of a network: a piece of hardware or software that decides which traffic may pass. As companies attach the word "AI" to products working at different layers, the term now covers at least three distinct jobs: hardware guarding a carrier or corporate network, a control layer guarding one company’s own AI application, and software guarding a single device. The three are not interchangeable, though vendors increasingly use the same label for all of them. ## Findings ### AI inside a company’s network firewall: Huawei Huawei unveiled AIFW at HUAWEI CONNECT 2026 in Shanghai. According to Antara News, the white paper says AIFW "integrates AI-native hardware, LLMs, and intelligence and security knowledge bases to shift defense focus from traffic protection to behavior governance," moving away from fixed rules toward what the paper calls policy autonomy. The risks it names include prompt injection, sensitive data leakage, compute misuse and unauthorised content leaving the network. This is carrier- and enterprise-grade hardware, bought and run by a company’s network team, guarding the edge of a corporate or telecom network. It is not something an individual installs on a laptop [[1]](https://en.antaranews.com/news/433177/huawei-releases-an-aifw-technical-white-paper-to-facilitate-firewall-upgrade). ### A firewall in front of a company’s own AI apps: Fastly Fastly's announcement covers a different layer again. AI Runtime Control is, in Fastly's own description, a control plane that sits between a company's applications and the AI model providers they call, routing every request through one place so a business can see and manage its AI usage: who is calling which model, and what it costs. The AI Firewall is a feature inside that control plane, built on Fastly's existing web application firewall, that inspects the natural-language traffic going to and from an AI model for attacks such as instructions hidden in a web page or a support ticket that try to hijack a customer-facing chatbot. [Fastly says](https://www.fastly.com/blog/govern-protect-and-control-every-ai-request-with-fastly) this runs at the network edge, without specialised hardware. Like Huawei's AIFW, this is a tool for the company building the AI application, not for the person using it. ### An on-device firewall that explains a Mac’s own connections: individuals FireAI uses the phrase "AI firewall" for a third, unrelated job: protecting one person’s own computer, not a company’s network or a company’s AI product. It runs as an Apple Network Extension content filter on a Mac, pausing each app’s connection and identifying the app by its code signature before anything leaves. An on-device AI model then writes a plain sentence describing what it found, without sending anything about those connections to a cloud service. This does not compete with Huawei’s or Fastly’s products; they guard a company’s infrastructure, FireAI guards the device in front of a user. ## Implications for Mac users and IT teams Direct relevance is limited to anyone working in IT for a company that runs Huawei network hardware or builds products on Fastly, in which case this is a matter for a security or platform team to evaluate. More broadly, "AI firewall" is now a label attached to at least three unrelated categories of product, and more vendors are likely to adopt the same phrase. Reading a label without asking what it actually guards is the habit that leads to a false sense of protection. > FireAI, the on-device firewall for macOS developed by HisnLabs, explains each app’s connections in plain sentences, generated on the Mac itself. A 17-day trial is available. [Download FireAI for Mac](https://hisnlabs.com/en/download) ## Recommendations 1. When "AI firewall" appears in a headline or an advert, ask what it actually sits in front of: a company network, a company’s AI application, or an individual device. The answer changes everything else. 2. For a tool claiming to protect a personal device, check where its AI processing runs. If connection history has to leave the device to be analysed, that is a trade-off worth knowing before installing it. 3. Do not assume "AI" in a product name means comprehensive protection. Check what it actually does: filter, explain, or only log, and read the product’s documentation rather than its tagline. 4. Evaluating Huawei’s or Fastly’s claims for a company network or AI application is a task for a security team working from the vendor’s own technical documentation, not from a headline. ## Relevance to FireAI FireAI is the third kind of "AI firewall" in this story: built for one Mac, not a data centre or an AI product. - Every connection an app makes is paused by FireAI’s [network filter](https://hisnlabs.com/en/docs/how-the-network-filter-works) and checked against [per-app rules](https://hisnlabs.com/en/docs/per-app-rules), with the app identified by its code signature so a fake copy cannot inherit a trusted app’s permissions. - FireAI’s [on-device AI model](https://hisnlabs.com/en/docs/on-device-ai-model) turns the facts it gathers into a plain sentence about what an app is doing, and stays on the Mac to do it: nothing about a user’s connections is sent to HisnLabs or to any cloud service. - [Security modes](https://hisnlabs.com/en/docs/security-modes), from Home to Under attack, let a user tighten what is allowed depending on where they are, such as on a public network. FireAI is not an enterprise network appliance like Huawei’s AIFW, and it does not sit inside a company’s infrastructure filtering prompts to a chatbot the way Fastly’s AI Firewall does. It is not an antivirus, so it does not scan files. It watches and explains the connections a Mac makes, and lets its user decide. FireAI is made by HisnLabs. > FireAI’s per-app rules and on-device AI model are built for one Mac, not a company network. A 17-day trial is available. [Download FireAI for Mac](https://hisnlabs.com/en/download) ## Limitations The account of AIFW rests on Antara News’s summary of Huawei’s own white paper; none of the sources here include independent testing of AIFW’s "policy autonomy" claims against real traffic. Fastly’s description of its AI Firewall’s effectiveness against prompt injection comes from Fastly’s own blog posts, with no independent security research cited in these sources. Neither product’s real-world performance has been verified by a third party in the reporting reviewed here. Try [FireAI, by HisnLabs](https://hisnlabs.com/en/download) free for 17 days. ## Sources - [Antara News, 27 September 2026: Huawei releases an AIFW technical white paper to facilitate firewall upgrade](https://en.antaranews.com/news/433177/huawei-releases-an-aifw-technical-white-paper-to-facilitate-firewall-upgrade) - [Fastly (Jonathan Speek, Terri Allegretto), 21 September 2026: Introducing Fastly’s AI Runtime Control: why your AI strategy needs a control plane](https://www.fastly.com/blog/introducing-fastlys-ai-runtime-control-why-your-ai-strategy-needs-control-plane) - [Fastly, 21 September 2026: Govern, protect and control every AI request with Fastly](https://www.fastly.com/blog/govern-protect-and-control-every-ai-request-with-fastly)