The Firewall Grew Up. Now Check Point Wants It to Police AI

In the beginning, Check Point could not afford the thing it planned to secure. An internet connection for the three-person startup cost about $1,000 a month in 1993, co-founder Gil Shwed later recalled, and they declined it. This was not visionary minimalism. It was a young company watching the cash. The founders still believed businesses would connect their private networks to the unruly public one - and would need a gatekeeper that understood more than a single packet at a time.

That insight became stateful inspection. Rather than treating each packet like a stranger at the door, the firewall remembered the conversation: where it began, where it was headed and whether the traffic made sense in context. FireWall-1 wrapped the technique in a graphical policy editor. The deep invention was memory; the commercial trick was making it usable by someone who did not enjoy typing rules into a terminal.

Thirty-three years later, the packets have company. Prompts, model calls and autonomous-agent actions now cross corporate networks. Check Point's new proposition is strikingly consistent with the old one: put intelligence at the point traffic already passes, remember enough context to judge it, and give the administrator one place to make policy. The nouns changed. The architecture kept its job.

A security company before security had a budget

The first thing that failed was urgency. In 1994 the web held only a few thousand sites; most businesses had no reason to care about internet threats because most were barely on the internet. Check Point could have had the right product several years too early. It survived the wait with roughly $400,000 from BRM Technologies, then used partners to carry the software into rooms the founders could not reach alone. Sun Microsystems signed an OEM deal in 1994. Hewlett-Packard followed with distribution in 1995.

This is the copyable part, with one caveat. A technical wedge travels farther when it fits someone else's route to market. Check Point sold through distributors, resellers and service providers instead of building every customer relationship itself. Its 2025 filing still describes a two-tier channel. The arrangement compounds: partners learn the product, customers accumulate policies and certifications, and renewals become less dramatic than replacement.

By the end of 1997, annual revenue had climbed from $9.5 million two years earlier to $82.9 million, with 14 straight profitable quarters. The company had found a rare startup combination: new category, legible interface, indirect distribution and software margins. Rivals could copy features. Reproducing the installed policy base and partner muscle was slower.

“You don't pick your battles, they pick you.”Gil Shwed, co-founder and executive chairman

Four pillars, fewer brand-name gymnastics

Check Point is no longer merely a firewall vendor, although the firewall remains its cultural center of gravity. The current portfolio is organized around four jobs. Hybrid Mesh Network Security covers physical and virtual firewalls, cloud networks, SASE, SD-WAN and web applications. Workspace Security protects email, collaboration tools, browsers, endpoints and mobile devices. Exposure Management finds weaknesses, adds threat context, validates what is exploitable and tries to remediate safely. AI Security discovers AI use and protects models, applications and agents from prompt injection, data leakage and abuse.

The customer is usually an organization with too much infrastructure and too few people willing to babysit it: a global enterprise, a government, a service provider, a midsize business or the managed-security firm acting on its behalf. Public examples range from Dropbox and Xero to airlines, banks, school districts and healthcare organizations. The problems are practical - phishing mail, ransomware, exposed cloud workloads, promiscuous firewall rules, rogue SaaS, vulnerable APIs and now employees pasting confidential material into AI tools.

Check Point's differentiation is consolidation with an old bias: prevention before cleanup. Competitors such as Palo Alto Networks, Fortinet, Cisco, Microsoft and CrowdStrike also bundle broad platforms; Zscaler and Netskope press from cloud-delivered access; specialist vendors attack every individual layer. Check Point argues that shared threat intelligence and policy across network, cloud and workspace can block more attacks while reducing consoles. The claim becomes valuable only when products genuinely share telemetry and action. A bundle is not a platform merely because procurement receives one invoice.

The revenue mix shows what changed management's mind about the future. In 2025, security subscriptions generated $1.219 billion and software updates and maintenance another $958 million. Products and licenses contributed $548 million. In other words, roughly four dollars in five came from recurring security and upkeep rather than the initial sale. In the second quarter of 2026, total revenue grew just 1% to $674 million, while security subscriptions grew 12% to $333 million. The mature core pays handsomely. The subscription layer has to supply motion.

Can the firewall understand an agent?

The July 2026 AI Network Firewall is Check Point's most literal attempt to turn heritage into leverage. Delivered in firewall software release R82.20, it is designed to discover sanctioned and shadow AI tools, inspect prompt intent, spot sensitive data leaving the network, identify Model Context Protocol traffic and block adversarial prompts before they reach an application. Existing enforcement points become AI checkpoints without a separate virtual appliance alongside them.

That is appealing because deployment cost is often the assassin of good security ideas. If an organization already routes branch, data-center or cloud traffic through Check Point, adding a software capability is easier than re-architecting the estate. Its agentic network-security tools make the same convenience argument for operations: administrators state business intent; agents propose or execute policy changes, tighten excessive access, troubleshoot failures and map configurations to compliance frameworks. High-impact changes retain human approval and an execution trail.

The limit is topology. This approach works when traffic passes through an enforcement point with enough visibility, decryption authority and identity context. It works less neatly when employees go directly to cloud services, unmanaged devices bypass controls, encrypted sessions remain opaque, or autonomous tools act entirely inside a SaaS provider. Endpoint, browser, API and cloud integrations then matter as much as the box. Check Point knows this; its “AI Defense Plane” stretches across those surfaces. Customers still have to deploy the pieces.

The other risk sits in the shopping cart. Check Point bought Perimeter 81 for about $490 million to fill out SASE, then added external-risk intelligence through Cyberint, exposure remediation through Veriti and AI runtime security through Lakera. Veriti alone cost approximately $92.5 million; Lakera's price was not disclosed. Acquisitions accelerate missing capabilities, but every one imports a data model, user experience and sales motion. Customers do not experience strategy slides. They experience whether alerts connect and policies propagate.

Nadav Zafrir inherited this integration problem when he succeeded Shwed as CEO in late 2024. Since then, the company has simplified the public portfolio into four pillars, pushed agentic management, expanded sales capacity and reframed its mission around securing enterprise AI transformation. Partnerships with AWS, Microsoft, Google Cloud, NVIDIA, Wiz, Illumio and OpenAI widen the surface it can cover without pretending to own every layer.

The old idea is still doing new work

Check Point occupies a peculiar market position: old enough to have invented one of cybersecurity's durable concepts, profitable enough to avoid theatrical reinvention, and mature enough that 1% quarterly growth invites hard questions. Its $2.6 billion in remaining performance obligations gives management time. Its 100,000-plus organizations give new products somewhere to land. Neither guarantees that firewall customers will buy AI governance from their firewall vendor.

But the company's best argument is refreshingly concrete. AI use is becoming network traffic. Agents create actions that require identity, policy and audit. Security teams already have too many tools. If Check Point can make the existing control plane see this new activity - and make acquired products behave like native organs - the firewall will not look like a relic. It will look like the same old bouncer, finally learning the names of the new guests.