“Blackstone is placing an early bet on Huskeys, a cybersecurity startup building an artificial intelligence gatekeeper to protect companies from the growing chaos of automated web traffic.
The firm's early-stage venture-capital arm, Blackstone Innovations Investments, led a $27 million Series A investment that values the year-old company at more than $100 million.
The deal makes Blackstone the second-largest outside investor in Huskeys, behind Israeli venture-capital firm 10D, according to a Huskeys representative. So far, investors have put $35 million into the startup.
Other backers in the most recent transaction include the investment arm of publicly traded cybersecurity company Zscaler and Bright Pixel Capital, the venture arm of Portuguese retailer Sonae.
New York-based Huskeys was founded in Tel Aviv in 2025 by Itai Gafni and Roy Weisfeld, veterans of the Israeli army's elite Unit 8200 cybersecurity division. Gafni is Huskeys' chief executive and Weisfeld is its chief technology officer.
Huskeys -- named for the founders' love of husky dogs and a play on the "keys" required to secure the web -- acts as an intelligent brain for cybersecurity systems. Its agentic AI layer sits on top of clients' current technology, "analyzing the logic and the business context behind every application" to spot threats that older technology might miss, Gafni said.
"AI is changing the game," he said, adding that the bulk of online traffic now is driven by AI agents, as opposed to people, making it harder to spot bad traffic. He compares what Huskeys does for web security to retrofitting an old car.
"I'm just replacing the engine, and now you have the most advanced car in the world," Gafni said.
The number of AI-driven malicious breaches rose by about 56% in the year through February, with AI-driven attacks driving up costs by about $1 million on average, according to a recent report from International Business Machines.
AI tools like Anthropic's Mythos are finding software bugs at an unprecedented rate, what security researchers call "Bugmageddon," allowing criminals to exploit vulnerabilities faster.
Security misconfigurations, such as the one the Justice Department highlighted in Capital One Financial's 2019 massive data theft, remain one of the top causes of data breaches.
Blackstone Senior Managing Director Adam Fletcher said companies can no longer rely on single-layer cybersecurity defenses and must instead rely on multiple layers of compensating controls, where even if one layer fails, other layers step in.
Overly restrictive firewalls can block real customers, costing companies money.
But letting too much traffic in poses a reputational risk as criminals could exploit vulnerabilities to access confidential data, said Fletcher, who led the investment with David Kim, a vice president at Blackstone Innovations Investments.
Huskeys provides an essential gatekeeper layer that addresses that, Fletcher said, citing Blackstone's own investor portal as an example: Blackstone needs to keep its website accessible to legitimate users while protecting confidential data.
Blackstone's early venture-stage investment group has backed such businesses as financial-technology companies iCapital, valued at more than $7.5 billion as of last year, and Canoe Intelligence, which is being acquired by Bloomberg.
Since Huskeys started doing business in early 2026, its revenue has roughly quadrupled quarter-on-quarter while customer growth has roughly quintupled, the company said.
The company plans to build a sales and marketing presence in the U.S. and has roughly doubled the number of employees in the last two months to nearly 40, said Gafni. Huskeys generates annual contract revenue in a six-figure dollar range, which gives investors an indication of recurring subscription revenue over 12 months, according to a company representative.
Companies that sell directly to consumers, like TikTok, account for most of its revenue, Gafni said. But the intelligent brain, he said, can be implemented across industries, from healthcare to energy. "This is a problem that everyone has," Fletcher said. "It's something that really has been woefully lacking innovation for a long time."
He added that Huskeys offers "an enhanced layer of defense against the types of attacks that we believe are coming."
---
Maria Armental writes for WSJ Pro Private Equity.” [A]
Let's compare Huskeys, a cybersecurity startup, with closest Chinese open-weights competitors on capabilities and price
Executive Summary
When comparing the Israeli cybersecurity startup Huskeys against Chinese open-weights AI models like Z.ai's GLM 5.3 and Moonshot's Kimi K3, you are looking at two entirely different architectural paradigms for cybersecurity.
• Huskeys is a specialized, enterprise-grade Network Edge Security Management (NESM) platform that adds an active AI gatekeeper layer to existing Web Application Firewall (WAF) infrastructure. It is purpose-built to block automated, malicious AI traffic in real time.
• Chinese Open-Weights Competitors are massive, general-purpose foundational LLMs that possess state-of-the-art software vulnerability, coding, and cyber-offensive capabilities. They are downloadable binaries that can be fine-tuned or hosted locally.
________________________________________
Direct Comparison: Huskeys vs. Chinese Open-Weights
Metric / Dimension Huskeys (Startup Platform) Chinese Open-Weights (GLM 5.3 / Kimi K3)
Primary Architecture Closed-source enterprise SaaS; Agentic AI layer built on top of CDN/WAF. Open-weight foundation models (downloadable mathematical parameters).
Core Cyber Focus Network defense & traffic triage: Differentiating between human, automated bad actors, and legitimate AI agents. Vulnerability discovery & code analysis: Scanning codebases for bugs, writing patches, or crafting exploits.
Real-time Edge Mitigation Native. Blocks malicious queries, handles orchestration, and deploys virtual patching dynamically. Requires custom pipeline. Models cannot block traffic natively; they must be wrapped in custom security pipelines.
Inference/Running Cost Predictable enterprise subscription/SaaS model tied to traffic/scale. 60% to 90% cheaper token cost than US frontier models, but requires hosting infrastructure.
Compliance & Jurisdiction Secured under Western venture backing (Blackstone, Zscaler, Okta). Potential compliance/regulatory risk; prompts can fall under Chinese data jurisdiction if using hosted APIs.
________________________________________
Capabilities Deep Dive
1. Huskeys: The Real-Time Traffic Gatekeeper
Huskeys operates at the network edge. Founded by veterans of Israel’s elite Unit 8200, the startup's platform addresses a massive upcoming shift: a world where up to 70% of internet traffic is driven by AI agents rather than humans.
• Autonomous Edge Defense: Instead of looking for traditional static hacking signatures, Huskeys uses an active AI layer to judge the intent of automated network traffic.
• Virtual Patching & Orchestration: It continuously rewrites WAF policies in real time to neutralize novel, automated attacks without breaking legitimate business logic.
• Enterprise Validation: Major firms like TikTok, Legoland, and Hugging Face actively use Huskeys to process trillions of web requests.
2. Chinese Open Weights: The Raw Analytic Engines
Models like Z.ai's GLM 5.3 and Moonshot’s Kimi K3 are raw reasoning engines. They do not passively monitor networks; instead, they serve as highly capable analysts or automated hackers.
• Uncensored Security Incident Resolution: Western frontier closed models (like OpenAI or Anthropic) have rigid safety guardrails that often cause them to refuse cyber-forensic tasks out of fear of assisting a hacker.
Notably, during a major hack on Hugging Face, the security team used GLM-5.2 because it didn't refuse to parse the breach data.
• Advanced Cyber-Reasoning: Benchmarks show models like Kimi K3 can successfully execute autonomous cyber campaigns—adapting exploits to unique environments and validating attack steps—trailing Western closed giants (like Anthropic's Mythos) by only about 6 months.
________________________________________
Pricing & Deployment Economics
• Huskeys Pricing: Backed by a $27M Series A led by Blackstone, Huskeys sells a full product ecosystem. You pay for the integration, real-time protection, and lower operational overhead (reduced false positives for security teams). You do not need to manage servers to host the AI.
• Chinese Open Weights Pricing: They are fundamentally disrupting market economics. If accessed via hosted APIs, Chinese models run 60% to 90% cheaper than OpenAI or Anthropic equivalents. If used as open-weights, the software license is free, meaning your only expenses are the internal infrastructure (GPU compute clusters) and the engineering talent required to build security tools around them.
________________________________________
The Strategic Choice
If your goal is to protect an enterprise application from being scraped or attacked by rogue AI bots, Huskeys provides an out-of-the-box, compliant solution backed by top-tier venture security names.
If your goal is to build a custom internal tool to audit software code, find zero-day vulnerabilities, or operate an internal red-team/blue-team sandbox, deploying a Chinese open-weight model locally provides near-frontier capabilities at a fraction of the cost—provided your corporate compliance policy permits it.
Has anybody designed open weight's agent, based on Chinese open-weights AI models like Z.ai's GLM 5.3 or Moonshot's Kimi K3, agent, that is able to do what Huskeys does?
Yes, developers are pairing models like GLM-5.3 and Kimi K3 with open-weight agent harnesses to tackle long-horizon tasks, though duplicating Huskeys' specific network-edge security management (NESM) for blocking automated AI traffic requires specialized setups.
Open-Weight Agents and Capabilities
• GLM-5.3 Orchestration: Z.ai’s GLM-5.3 is frequently coupled with the Z-code harness for autonomous coding and multi-step cybersecurity exploitation tasks.
• Kimi K3 Orchestration: Moonshot AI's Kimi K3 features a 1-million-token context window and Kimi Delta Attention, making it popular for multi-tool procedure execution and repository navigation.
• The Huskeys Gap: While these open-weight models match frontier reasoning, running a 2.8-trillion-parameter model like K3 locally for real-time edge filtering requires massive enterprise infrastructure (hardware purchase: $1.9 million to $3,000,000 upfront for the rack infrastructure. operational costs: around $140,000+ per year purely in electricity and cooling infrastructure overhead (drawing roughly 132 kW per rack)), meaning most security agent setups rely on hosted APIs rather than fully local self-hosting. [1]
Self-hosting protects you from Huskeys spying on your trade secrets. This is worth investing in, if you can afford it. Important, that you don’t have to pay anybody for tokens.
A. Blackstone Invests in Cyber Startup. Armental, Maria. Wall Street Journal, Eastern edition; New York, N.Y.. 03 Sep 2026: B4.
Komentarų nėra:
Rašyti komentarą