Israeli start-up Zeroport seeks to put physical limits on AI agents

ByANNA AHRONHEIMNOVEMBER 9, 2026 10:08

As companies increasingly integrate artificial intelligence into their operations, Israeli start-up Zeroport believes that the next major cybersecurity challenge will be keeping autonomous AI systems under control.

Founded in 2024 by CEO Joseph Gertz and CTO Lavi Friedman, Zeroport has developed a hardware-based security platform designed to sit between AI systems and enterprise networks, creating a physical layer that can block unauthorized activity before it reaches critical infrastructure.

The company believes that traditional cybersecurity solutions were designed primarily to protect against human users and external attackers. The growing use of AI agents capable of independently accessing systems, processing information, and executing tasks is creating a different security challenge.

"The security industry has spent decades building software to defend against software," Gertz told The Jerusalem Post. "But AI changes the equation. If these systems become capable enough, then software-based controls are no longer enough.”

Gertz recounted that, like many other start-ups over the past three years, the idea emerged during reserve military service in May 2024, when he and Friedman began discussing the implications of rapidly advancing AI capabilities after they identified a potential gap in existing security systems.

Their approach was to move security enforcement from software into physical infrastructure.

“We saw that we needed to build a solution that brings together something physical, whereas other companies are trying to do that with software,” Gertz said. With their technology, "whatever AI tries to do, it will meet a physical wall and not just software that it can get around.”

He explained that at the center of Zeroport's technology is a proprietary hardware component built around custom-designed chips and circuit boards developed by a team of electrical engineers and hardware specialists. The devices are installed at the boundary between cloud environments, data centers, and enterprise networks. Acting as an independent security checkpoint, they monitor activity crossing that boundary and can physically interrupt connections when an AI system attempts to perform an unauthorized action.

Rather than relying exclusively on software instructions to stop an agent, Zeroport's system is designed to enforce restrictions at the hardware level. Zeroport’s hardware is positioned between the AI environment and the organization's internal systems.

Gertz compares the technology to a drawbridge protecting a castle, telling the Post that "if an anomaly is detected, a drawbridge goes up. The connection is physically interrupted."

It’s “like a physical kill switch any time an agent does something you don’t want it to do,” he said.

The technology is intended to address concerns surrounding agentic AI, where systems are capable of independently making decisions and carrying out tasks with limited human supervision. And with companies increasingly relying on software-based technology to restrict AI activity, Zeroport believes that those protections may become insufficient as AI systems grow more capable.

Gertz explained that one emerging concern is what is known as "agentic drift," in which an AI agent receives a legitimate instruction but gradually expands its activities beyond the original task. For example, an agent initially authorized to analyze a company's network could attempt to access additional systems, information, or workflows that were not part of its original task.

The company believes enterprises will increasingly face a choice between allowing external AI agents to operate within their networks and building their own AI infrastructure, including dedicated computing capacity.

Preparing for the quantum age

Building a hardware-based cybersecurity company presents different challenges from developing a software product, particularly in terms of engineering, manufacturing, and investment. But Gertz believes that as enterprises deploy increasingly powerful AI models, security mechanisms must be capable of operating independently of the systems they are designed to control.

“Building the layer of hardware enables you to make sure you are the one that makes sure the agent does what you want it to,” he said, adding that “software is the brain.”

The emergence of sufficiently powerful quantum computers could undermine many of the encryption protocols currently used to secure digital communications, creating additional challenges for companies managing sensitive data and infrastructure.

Gertz said Zeroport has developed a symmetric encryption protocol that does not rely on transmission over the Internet and is incorporating encryption designed to withstand future quantum threats. That would allow the company to make sure that its products remain relevant as quantum computing capabilities advance over the coming decade.

The start-up says it has already secured over a dozen customers around the world, including data centers, Fortune 1000 companies, and organizations building AI computing infrastructure. Zeroport has also raised backing from investors including Lool Ventures, Fusion Fund, and Clarim Ventures.

Follow us on Google

---