Exein Funding Round Bets $270 Million on Fighting AI Hackers With AI

Exein raised $270 million after detecting about 5,000 new attacks each week across its network, five times the volume reported one year earlier. The Exein funding round values the Italian cybersecurity company at $1.7 billion. It also backs an ambitious wager: attackers using AI against connected machines will require defensive AI operating inside those machines.

That is a bigger claim than another cybersecurity company adding an AI assistant. Exein wants to protect robots, vehicles, industrial systems, and medical equipment at the firmware and operating-system level. These systems can cause physical consequences when compromised, rather than simply exposing information.

The company now faces a difficult test. It must convert proprietary device telemetry into autonomous defenses without introducing new errors, delays, or unacceptable control risks. Meanwhile, established security vendors such as Armis, Claroty, and Dragos already defend connected environments through asset discovery, network monitoring, and threat intelligence.

The Exein Funding Round Turns Device Security Into a Scale Contest

The new capital gives Exein the resources to compete for a security position inside connected machines before larger vendors close the gap.

Exein announced the financing on September 15, 2026. Headline led the round, with Sofina, Goldman Sachs, the European Investment Bank Group, KfW Capital, and T.Capital participating.

Existing investors also returned. They include Balderton Capital, HV Capital, Intrepid Growth Partners, 33N Ventures, Lakestar, Supernova Invest, Blue Cloud Ventures, and Geodesic Capital.

The company said the round was significantly oversubscribed. Separately, Exein expanded a revolving credit facility led by JPMorgan, with KfW joining as another lender. That facility gives Exein additional capacity for acquisitions in Europe and the United States.

The financing follows two large rounds completed during 2025. Exein raised €70 million in July and another €100 million in December. The company said the later investment brought its total 2025 financing to €170 million.

This latest transaction carries a $1.7 billion valuation. That places Exein among Europe’s most valuable privately held cybersecurity companies. Axios described it as Europe’s second most valuable cybersecurity start-up, behind Nord Security, in its funding coverage.

The valuation reflects more than revenue expectations. It represents a bet that security spending will migrate closer to the devices controlling physical processes.

Exein says its technology already reaches more than two billion devices. That figure describes the company’s technology footprint, not necessarily two billion separately billed installations. The distinction matters when judging commercial scale.

The company also says it observes roughly 5,000 new, non-repetitive attacks each week across that network. It reports that this volume has increased fivefold within one year.

Those figures have not been independently audited in the public materials. However, they explain why investors view distribution as Exein’s central advantage. A security provider installed across many device types can collect behavioral signals that a smaller laboratory deployment cannot reproduce.

Exein plans to use the money for expansion in the United States and Asia-Pacific. It has established a regional headquarters in Taiwan and a legal entity in Japan. The company expects Japanese expansion during the fourth quarter of 2026 and plans a South Korean office by 2027.

A Bay Area office is also planned. That move puts Exein closer to American chipmakers, cloud providers, robotics developers, and manufacturers. It also places the company directly against better-funded security incumbents.

The transaction therefore changes the competitive clock. Exein no longer needs to prove only that embedded protection works. It must show that its approach can become a repeatable global product before rival platforms extend deeper into firmware.

Why AI Hackers Are Moving Below the Network

AI increases the speed of finding and exploiting device weaknesses, while connected machines give successful attacks consequences beyond stolen data.

Traditional enterprise security often concentrates on laptops, servers, cloud accounts, networks, and employee identities. Connected machines present a different problem. They use specialized hardware, long-lived firmware, constrained computing resources, and operating environments that cannot tolerate routine downtime.

A vulnerability in a business application can expose records. A vulnerability in a machine controller can stop a production line, alter a vehicle’s behavior, or interrupt medical equipment.

That difference explains Exein’s use of the term physical AI. It refers to AI systems that sense, decide, and act through machines in the physical environment. Robots, drones, vehicles, and industrial controllers fit this category.

Attackers do not need a malicious general-purpose robot model to create harm. They can target the operating system, firmware, software dependencies, update process, or credentials surrounding the AI system.

Generative AI reduces some barriers to those attacks. It can help adversaries examine code, translate technical documentation, produce exploit variations, and automate reconnaissance. Human expertise remains important, but one attacker can test more paths in less time.

Academic research is already examining this shift. A 2026 paper on consumer robot security reported that an automated cybersecurity agent found 38 vulnerabilities across tested robotic platforms. The result does not establish that every robot faces the same exposure. It does show how AI can compress specialized security work.

Connected-device defenses also face an architectural disadvantage. Many organizations cannot install conventional endpoint tools on every controller or embedded system. Some devices have limited memory and processing power. Others must remain available for years without major software changes.

Network monitoring helps security teams identify suspicious communication. Asset-discovery platforms help them determine which devices exist and which vulnerabilities affect them. Neither method automatically stops malicious code executing inside a device.

Exein argues that protection should live closer to that execution point. Its software is designed for integration into device firmware, allowing security controls to observe behavior from within the machine.

That position matters because network boundaries keep changing. A connected camera, industrial gateway, or robot can communicate with cloud services, local controllers, maintenance tools, and third-party applications. A perimeter product may see traffic without understanding every action occurring inside the device.

The European regulatory calendar adds urgency. The European Union’s Cyber Resilience Act applies cybersecurity requirements to products containing digital elements. It places responsibilities on manufacturers across design, development, production, and vulnerability handling.

The law’s incident-reporting requirements took effect on September 11, 2026, four days before Exein announced its financing. Its main product-security obligations become applicable on December 11, 2027.

Manufacturers must report qualifying actively exploited vulnerabilities and severe incidents through a single European platform. The reporting framework includes an early warning deadline of 24 hours and a fuller notification within 72 hours.

This regulation does not require manufacturers to buy Exein’s products. It does increase the value of knowing what software is running, detecting abnormal execution, and responding quickly when exploitation begins.

For hardware manufacturers, cybersecurity is moving from an optional enterprise layer into the product lifecycle. That creates an opening for embedded vendors. It also invites larger security companies, semiconductor suppliers, and operating-system providers to offer competing controls.

Exein’s AI Defense Starts Inside the Machine

Exein’s strategy combines kernel-level runtime enforcement with a planned foundation model trained on machine behavior rather than human language.

The first component is runtime security. A runtime system observes or controls software while that software is executing. This differs from scanning firmware before release or reviewing network traffic after a device starts communicating.

Exein introduced an architecture called Photon earlier in 2026. The company says Photon operates at the kernel level, the privileged layer that manages processes, memory, devices, and system permissions.

Photon is designed to stop malicious execution before it completes. Exein has not publicly released enough independent performance data to establish how consistently it achieves that result across every supported environment.

The architecture builds upon Exein’s earlier work with eBPF, a Linux technology that runs verified programs inside the kernel. Exein’s open-source Pulsar project uses eBPF to collect and analyze system activity without requiring conventional kernel modules.

This approach offers an important advantage. Security logic can observe processes and system calls close to their source. It can therefore identify behavior that a remote monitoring system might miss.

Yet proximity also raises the standard for reliability. Security software operating near the kernel must avoid crashes, excessive resource use, and false positives. A mistaken block on a laptop is disruptive. The same error inside a vehicle or hospital device can be more serious.

The second component is Exein’s planned AI model. The company says it is training a proprietary foundation model using two years of telemetry from its device footprint.

A foundation model learns broad patterns from large datasets and can support more specialized tasks. Exein’s proposed model focuses on machine behavior rather than ordinary text, according to the company’s funding announcement.

That dataset is the heart of the investment thesis. A model trained on real operating behavior could learn which process combinations, resource changes, and execution sequences usually indicate danger.

The defensive agent could then react locally, without waiting for a human analyst to review an alert. That speed would directly address attacks generated or adapted by offensive AI.

Exein expects its agentic security architecture to ship by the end of 2026. It expects the first foundation models during the first quarter of 2027.

Those are forward-looking company targets, not completed product milestones. They also leave several technical questions unanswered.

The company has not publicly specified the model’s architecture, evaluation benchmarks, or decision boundaries. It has not detailed how the model handles unfamiliar devices, rare legitimate behavior, or attempts to manipulate its observations.

Data quality presents another challenge. A footprint of two billion devices sounds defensible, but scale alone does not guarantee representative training data. Different machines use different processors, operating systems, workloads, and safety constraints.

A model trained heavily on cameras or industrial gateways might not transfer safely to autonomous vehicles. Even similar devices can behave differently across customers, regions, and software versions.

Telemetry also creates privacy and governance questions. Exein says its model uses operational machine behavior rather than human-generated text. Buyers will still want precise answers about collection, retention, anonymization, regional storage, and access controls.

The strongest version of Exein’s approach does not let a central model freely control every machine. It combines broad learning with narrow, auditable policies suited to each device.

That design would preserve machine-speed detection while limiting autonomous actions. It would also help manufacturers document how defensive decisions occur.

For buyers, the practical question is not whether Exein uses AI. It is whether the system improves detection and response without increasing operational risk.

The Real Contest Is Embedded Defense Versus External Visibility

Exein is challenging the idea that organizations can secure physical systems mainly by discovering assets and monitoring their networks.

Security vendors such as Armis, Claroty, and Dragos have built large businesses around connected assets and operational technology. Their platforms generally emphasize visibility, exposure management, network behavior, industrial protocols, and threat intelligence.

These functions solve a serious problem. Many organizations do not know every connected device operating in their facilities. Security teams cannot protect assets they cannot identify.

Passive network monitoring is especially useful in operational environments. It can detect devices without installing software or actively scanning systems that might be fragile.

Exein starts from another position. It targets manufacturers and technology suppliers that can integrate protection into firmware before products reach customers. Its product travels with the device rather than depending entirely on the owner’s network controls.

Neither approach eliminates the other. An embedded runtime cannot provide a complete view across factories, hospitals, or transportation systems. A network platform cannot always observe the precise code path running inside a machine.

The central competition concerns which layer becomes the primary enforcement point. If attacks move faster and target execution below the network, Exein’s position gains value. If manufacturers resist embedded agents, external monitoring retains the operational advantage.

Integration creates the first hurdle. Device makers must evaluate compatibility, memory use, processor overhead, update behavior, and long-term maintenance. Each product family can require separate engineering and certification work.

Sales cycles create the second hurdle. A security platform sold to an enterprise can begin producing value after deployment. Firmware integrations may depend on product-development cycles lasting several years.

Exein can address this delay through partnerships with chipmakers and original equipment manufacturers. Integration at a common hardware or software layer can place its technology into many downstream products.

The company has previously announced relationships involving MediaTek and industrial-computing supplier Kontron. It also identifies organizations such as Nvidia, Arm, AWS, Intel, Hitachi Energy, MediaTek, and Daikin among companies using or working with its technology.

Those relationships vary in scope. A technical integration, marketplace listing, customer deployment, and strategic partnership do not carry identical commercial meaning. Public descriptions should not be treated as equivalent contracts.

Competition may also come from below. Semiconductor vendors increasingly include secure boot, trusted execution, cryptographic identity, and hardware isolation in their products. Operating-system providers can add their own runtime policies.

It may come from above as well. Large cloud and security platforms can combine device inventories, vulnerability intelligence, and AI-assisted response. Acquisitions could quickly move an established vendor deeper into embedded protection.

Exein’s expanded credit facility suggests management recognizes that pressure. Acquiring complementary technology or distribution could accelerate its move into markets where organic integrations would take too long.

Regulation strengthens both sides. The Cyber Resilience Act requires manufacturers to assess risk and manage vulnerabilities across supported product lifecycles. The official CRA summary also describes documentation, conformity, and reporting responsibilities.

Embedded controls can support those obligations by generating device-level evidence. External platforms can support them through asset inventories, vulnerability management, and incident workflows.

The winner will therefore not be chosen by a single AI benchmark. Buyers will evaluate coverage, deployment effort, auditability, resource use, and response reliability.

Exein’s funding gives it time to prove that firmware-level enforcement belongs in this stack. It does not guarantee that manufacturers will accept one provider across diverse product lines.

What the $1.7 Billion Valuation Still Has to Prove

Exein must validate three things next: safe autonomous response, repeatable manufacturer adoption, and expansion that produces measurable commercial scale.

The first signal will arrive with the planned agentic architecture. Exein says it intends to ship that system before the end of 2026.

A launch alone will not settle the question. Buyers need evaluation results showing detection quality, false-positive rates, resource consumption, and recovery behavior. They also need to know which actions remain under human control.

Successful deployments would strengthen Exein’s central argument. Delays or tightly limited capabilities would suggest that machine-speed autonomy remains harder than the funding narrative implies.

The second signal is the first foundation-model release, expected during the first quarter of 2027. Exein should explain how it evaluates the model across device classes and how it prevents unsafe generalization.

Independent testing would carry more weight than internal claims. Comparisons should include conventional rules, anomaly detection, and network-based systems. Buyers need evidence that the model adds value beyond familiar security techniques.

The model must also support investigation. Security teams cannot rely on unexplained blocking decisions when a machine controls a critical process. Useful output should connect an action to observable behavior and an enforceable policy.

A transparent evaluation would strengthen the company’s claim that proprietary telemetry creates a lasting advantage. Weak or narrowly scoped results would reduce the value assigned to that dataset.

The third signal is commercial adoption outside Europe. Exein plans new operations in the Bay Area, Japan, and South Korea while expanding its existing base in Taiwan.

Office openings reveal ambition, not product acceptance. More useful indicators include production integrations, deployed product families, renewed contracts, and growth within existing manufacturers.

The company also needs to clarify how its two-billion-device footprint translates into recurring revenue. Wide technical distribution can produce modest economics when licensing terms are small or indirect.

Manufacturers will examine another tradeoff. Installing autonomous defense inside a product creates a dependency on its security supplier. Buyers must understand update commitments, support periods, failure modes, and migration options.

The Cyber Resilience Act makes those questions harder to postpone. Its lifecycle obligations encourage manufacturers to choose controls they can maintain for years, not short-lived experiments.

Exein’s acquisition strategy deserves attention for the same reason. A carefully selected purchase could add firmware analysis, device identity, vulnerability management, or American distribution. An unfocused deal could complicate a platform already spanning many environments.

Investors are effectively financing a race between two accelerating systems. Attackers can use AI to scale discovery and exploitation. Defenders want AI to recognize and block that activity at comparable speed.

The symmetry is attractive but incomplete. Attackers can tolerate many failed attempts. Defensive software must remain accurate while protecting systems that cannot simply be restarted.

That imbalance is the strongest skeptical case against the Exein AI defense plan. Fast autonomous protection sounds appropriate for fast autonomous attacks. However, speed cannot replace verification when defensive errors carry physical consequences.

Exein can answer that concern through bounded autonomy, documented policies, independent testing, and narrowly defined responses. Its technology should stop clearly malicious behavior while escalating ambiguous cases.

The Exein funding round has bought the company geographic reach, product-development time, and acquisition capacity. It has also raised expectations far beyond those attached to a conventional embedded-security vendor.

Developers should watch the model evaluations and supported environments. Manufacturers should examine integration effort, evidence quality, and lifecycle commitments. Enterprise buyers should ask how embedded enforcement works alongside their existing asset and network platforms.

The decisive question is now concrete. Can Exein turn machine telemetry into dependable security decisions before AI-assisted attackers exploit the same machines at greater speed?

Watch the end-of-2026 architecture release first, then the planned model in early 2027. After that, look for named production deployments in the United States and Asia-Pacific. Those results will show whether the Exein funding round financed a durable security layer or an expensive race to prove one.