Sternum

Artificial intelligence is a monitoring layer beside a protection technology that uses none. Sternum's core product, EIV embedded integrity verification, instruments a device's compiled firmware with checks on execution flow and memory, which the company calls a "deterministic security model": it blocks or it does not, with no learned model involved.

The learned component sits in the observability platform: "When your device is connected, the AI will auto-train itself to learn its routines and use the information to alert of anomalous activities", described as multivariate analysis for IoT threats and an "extra pair of eyes" for operators. Runtime protection, forensics, remote debugging and log management stand without the AI.

The company claims its protection never fails, and its own figure says otherwise. The runtime protection page says EIV "categorically prevents all code and memory manipulation attempts and assures system integrity at all times" and "deploys verification checks across all exploitation paths to deterministically prevent all code and memory manipulation attempts"; the home page says it will "deterministically prevent all exploit attempts".

The same runtime protection page reports "96.5% Prevention in industry benchmark tests for memory vulnerabilities", without naming the benchmark. The protection acts on its own on the device, blocking what it judges a manipulation without a person in the loop, and every blocked attempt is logged with forensics. A manufacturer told that all exploits are prevented, and that patching can fall by up to 60 percent, has reason to slow the patching that would catch what EIV misses. Nothing published says what happens to device function when EIV blocks something legitimate, or which attack classes fall outside it.

The protection method is named and its cost stated; the anomaly model is not identified. EIV is described as binary instrumentation that auto-profiles firmware, including third party components, and places verification checks on execution flow and memory, integrated into the build, test and deployment process, with overhead given as 1 to 3 percent. Supported environments are named, including FreeRTOS, Zephyr, QNX, Micrium, Mbed OS, Azure ThreadX and embedded Linux.

The AI anomaly engine is described only as multivariate analysis that trains itself on each device's routines after a short period. No model, training data, version or change record is published for either layer.

Short at the model layer, open everywhere else. The anomaly engine is Sternum's own and trains itself on each device's data, so no outside model provider is described. The data does leave the device: the continuous monitoring product is a "Cloud platform" that uses a proprietary observability SDK to collect device-level data and fleet trends, which customers can also forward to their own SIEM or SOC through a REST API.

Sternum's privacy policy says personal data may be stored by Sternum, its affiliates and service providers in the United States, Israel, the European Union and other jurisdictions, and lists the service providers only by category (hosting, data analytics, development, support, security and others); none is named.

Named customers and partners, and figures without method. NXP Semiconductors' Marc Vauclair is quoted: "All attacks NXP crafted were blocked by Sternum with CPU overhead of less than 3%"; a former product security director at Medtronic, Telit's CISO, and Vibrant and HARDWARIO are quoted, and the Linux Foundation's Zephyr lead describes the partnership.

The site reports 96.5 percent prevention "in industry benchmark tests for memory vulnerabilities", patch management savings of 40 percent on the home page and up to 60 percent elsewhere, and more than 100 million events observed a day. The benchmark, NXP's test design and the savings baseline are not described, and no independent evaluation is published.

A privacy policy reaches the service, and it keeps room to reuse data. Sternum's privacy policy covers personal data of device end users that customers submit through the service, with Sternum acting as the customer's processor and keeping that data according to the customer's instructions; it says Sternum does not review submitted data and cannot classify it as personal or sensitive.

It also lets Sternum create aggregated, inferred non-personal or anonymized data "which we or our business partners may use to provide and improve our respective services", and transfer non-personal data at its discretion. The anomaly model trains on each device's own routines. No retention period, statement on training beyond the device, or de-identification method is published, and nothing covers how an AI alert that proves wrong is handled. On a medical device, telemetry could carry patient data, which the policy's no-review stance leaves to the manufacturer.

A privacy document reaches the service; no business associate position is taken. Sternum's privacy policy covers end user personal data that customers' devices send through the service, names Sternum as processor for the customer, and says data is kept according to the customer's instructions. It says nothing about protected health information or business associate agreements, and states that Sternum does not review submitted data and cannot tell whether it is sensitive. For a medical device manufacturer whose device telemetry may carry patient data, whether Sternum will sign a business associate agreement is not answered on the site.

Nothing independent is published about Sternum's own security. No SOC 2 report, ISO/IEC 27001 certificate, penetration test of the platform or trust center appears on the site; the privacy policy says only that industry-standard physical, procedural and electronic measures are used. NXP's test of EIV against crafted attacks concerns the product's protection, not how Sternum runs its own systems. The product sits inside the compiled code of the customer's device and its SDK sends device data to Sternum's cloud platform, so it becomes part of the device's software supply chain.

No device claim; the company works device regulation for manufacturers without stating its own position. Sternum is security software that manufacturers embed in their products, with no FDA clearance and none claimed. The home page says EIV can "serve as IDS/IPS for Medical Devices", the compliance page cites "Deep IoMT Expertise" with medical device manufacturers and offers device data as regulatory evidence, and the site runs a regulation learning center.

Once embedded, EIV becomes part of the device's software, so its behavior falls under the manufacturer's own submission and change control, and no page states how Sternum supports that or what intended purpose it states for EIV itself.

Nothing is published on how the anomaly model is governed or tested. The AI layer trains itself on each device's routines and then "can autonomously digest data at scale and instantly pinpoint critical insights and signs of abnormal behavior", but no evaluation, false alarm rate, framework or review process is described, and no result is given by device type, operating system or deployment environment, which is where a self-trained baseline is most likely to misread normal behavior. The deterministic EIV layer raises no model governance question of its own, and its benchmark figure is recorded under evidence.

Forensics let a manufacturer examine what the product did; nothing stands behind it. Every blocked attempt is reported with logs and attack path detail, and the observability platform adds remote debugging and root cause analysis, so a manufacturer can trace an EIV block or an AI alert. No terms, warranty, indemnity or remediation commitment are published, and the one efficacy figure, 96.5 percent prevention in an unnamed industry benchmark, comes without method or denominator.

It has no clinical record surface; its connections are to a manufacturer's device platforms and security tooling, and none of the latter is named. EIV integrates into the build, testing and deployment process and supports named embedded platforms (FreeRTOS, Zephyr, QNX, Micrium, Mbed OS, Azure ThreadX and embedded Linux), and the Linux Foundation describes Sternum as a Zephyr partner.

Monitoring data goes to Sternum's dashboards or "to your SOC, SIEM or SOAR" through a REST API, but no SIEM, SOAR, ticketing or build system is named and no production integration is shown. The operating systems are platforms the product runs on, not systems it exchanges data with.

Protection runs on the device; the monitoring platform is a single cloud service with no committed location. EIV executes inside the device's firmware and keeps working when the device is offline, with 1 to 3 percent overhead. The observability and threat detection products run on Sternum's cloud platform, fed by an SDK on the device, and data can be forwarded to the customer's own security tools.

The privacy policy says data may be stored in the United States, Israel, the European Union and other jurisdictions, which describes possibilities rather than a committed region, and no on-premises platform, region choice or tenant separation is stated.

No price, and a free evaluation route. Sternum offers a free evaluation kit: an ARM-based development board, a trial of EIV runtime protection and a trial account on the threat detection and observability platform, requested by form; an OpenWrt evaluation is marked coming soon. No price, license unit or tier is published, and nothing says whether licensing runs per device shipped, per product line or per platform.

Coverage is named, with validation behind part of it. Sternum sells to manufacturers of connected devices across medical, industrial, critical infrastructure and consumer IoT, new and legacy, on RTOS and embedded Linux, and presents EIV as IDS/IPS for medical devices. Named validation comes mostly from outside healthcare (NXP, Telit, Vibrant, HARDWARIO, the Zephyr Project); on the medical side a former Medtronic product security director is quoted. Medical devices are one vertical among several.