Armis

The model drives one core module of a wider platform. Armis describes a cloud threat detection engine that "uses machine learning and artificial intelligence to detect when a device is operating outside of its normal 'known good' baseline", comparing each asset's live behavior with baselines for similar assets seen in other customer environments.

The baselines come from the Asset Intelligence Engine, a crowdsourced knowledge base the company sizes at over five billion assets on its engine page, over six billion in a 2025 solution brief and over 6.5 billion on the medical device page. Behavioral detection does not exist without those learned baselines.

Discovery through integrations and traffic analysis, inventory, recall and advisory matching, CMMS work orders and policy enforcement carry value without the models, and Medical Device Security is one product on a platform also sold for IT, OT, IoT, cloud and application security, where a generative assistant and language model features sit beside it.

The medical device page makes an absolute detection claim with the system as its subject: "We identify and classify every device the moment it connects to your network, so nothing is left unmonitored." The same page says Armis "discovers and classifies every asset without needing to install any software agents" and delivers "total, real-time visibility and protection of every asset in the modern healthcare environment".

A buyer who takes that at its word has no reason to reconcile the inventory against CMMS or purchasing records, which is the step that would catch a device the platform missed or misclassified. The platform also acts on its own findings: "If a device exhibits abnormal behavior, its network access can be automatically restricted", and it "can automatically enforce policies to quarantine a suspicious device or remove its network access" through the customer's firewalls, switches and network access control.

Nothing on the medical device page, the platform terms or the trust center describes an approval step, a simulation of enforcement impact, a rule exempting a device in clinical use from automatic quarantine, or what happens after a wrong quarantine. The platform terms disclaim any warranty that the solutions will detect every threat or vulnerability.

The approach is described and some suppliers are named; versions and update practice are not. Armis describes behavioral detection that compares each asset with "known good" baselines for similar assets seen in other environments, built on a crowdsourced knowledge base that records how often each asset communicates, over which protocols, how much data it moves, whether it is stationary and what software it runs.

The platform terms of 14 May 2026 say anonymized, aggregated statistical data from customer use trains Armis's machine learning algorithms and artificial intelligence models. The subprocessor list dated 12 September 2026 names Amazon Web Services Bedrock as the language model inference provider for non-US customer tenants and for interactive product documentation, and Cerebras Systems as a language model inference provider for application security; it does not name the inference provider for US tenants.

No model, model version, accuracy figure or change record for the detection and classification models is published. A platform AI disclosure FAQ and an AI assistant privacy statement are listed in the trust center behind an access request.

A public subprocessor list names the model layer for part of the platform. The list dated 12 September 2026 gives each party, its service, the data it receives and its locations: Amazon Web Services for hosting and, through Bedrock, language model inference for non-US customer tenants; Cerebras Systems for language model inference on customer repository data in application security; Google for indicator of compromise classification on IP addresses, MAC addresses and domain names, in the United States; Cloudflare, Confluent and SingleStore on all customer data, in the region the customer selects; MongoDB and Sentry for vulnerability prioritization; and Atlassian, Coralogix, FullStory, Pendo, Salesforce, Twilio and Slack for support, logging, notifications and product analytics, with four Armis group entities providing support.

The list does not name the inference provider for US tenants, and no retention or training terms governing data at the model providers are published; the AI disclosure FAQ and assistant privacy statement may hold them, but both sit behind an access request in the trust center.

Named hospitals, one figure without method, and analyst recognition. In a June 2024 Armis release, Main Line Health reported finding twice as many connected assets as previously estimated, with no count, baseline or period given.

The medical device page quotes Main Line Health's system director of biomedical engineering and the CIO of Mater Misericordiae University Hospital, and a 2025 solution brief quotes University Health Network and Burke Rehab Hospital; none of these quotes carries a figure.

Armis reports being named a Leader in the 2026 Gartner Magic Quadrant for CPS Protection Platforms for a second consecutive year, an analyst assessment of the platform across industries rather than a measure of medical device identification or detection. No independent test of classification accuracy, detection rate or false positive rate is published.

Stewardship terms are published, and they grant the vendor use of customer data. The platform terms of 14 May 2026 give Armis the right to use customer data "as necessary to provide the Armis Solutions to Customer and to improve the Armis Solutions", and let it own and use statistical data, defined as data from the customer's use that has been "irreversibly anonymized as to Customer and aggregated", excluding IP addresses, MAC addresses and customer identifiers, for purposes that include "training Armis' machine learning algorithms and artificial intelligence models".

Customer data is deleted 60 days after termination on an automated schedule, with export available before then. No retention window during the term is stated, and the Asset Intelligence Engine page says the platform keeps "a history on everything each asset does". The medical device page offers monitoring for the transmission of unencrypted PHI, so what the collectors capture from a hospital network is a live question.

Nothing is published on guardrails for the AI assistant, how its output is checked, or how a safety event is handled. An AI assistant data privacy statement is listed in the trust center behind an access request.

A service privacy document exists and no business associate position is taken. The platform terms govern customer data, defined as the network, device and code data the platform collects, and commit to deleting it 60 days after termination; a data processing addendum dated 5 April 2024 is listed on the legal page, though the linked PDF is not available.

Neither those documents nor the trust center, which lists fifteen compliance programs with no HIPAA or HITRUST entry, states business associate status, offers an agreement, or says the service stays outside HIPAA. The medical device page markets monitoring for "the transmission of unencrypted PHI" and "streamlined HIPAA audit preparation" for the customer: that describes the customer's compliance rather than the vendor's own position, and it implies the collectors can observe traffic carrying patient data.

Recognized certifications are named in a public trust center, without the report type or dates on its open face. The Armis Trust Center lists SOC 2 with no type stated, ISO/IEC 27001 with its statement of applicability and no version, ISO/IEC 27017:2015, ISO/IEC 27018:2019, ISO/IEC 42001:2023, FedRAMP Moderate, Department of Defense Impact Level 5, TX-RAMP Level 2, BSI C5, CSA STAR Level 1, Cyber Essentials, the Spanish ENS scheme, TISAX and a VPAT, plus a penetration test report and its remediation report.

Documents are released through a "Get access" request, and the page does not say whether a self-serve agreement or an Armis approval releases them. Amazon Web Services is named as the infrastructure provider, with a separate production environment. The legal page adds a security advisories page, a vulnerability disclosure policy and a bug bounty program.

No device claim, and the product works with the device regime on the customer's behalf without stating its own position. Armis Centrix for Medical Device Security is security software with no FDA clearance and none claimed. The medical device page and a 2025 solution brief say it tracks FDA recalls and alerts, correlates recalls and security advisories with the affected devices, and correlates manufacturer MDS2 forms with observed risk; the page also offers HIPAA audit preparation for the customer.

The federal and state authorizations held for the cloud service (FedRAMP Moderate, Department of Defense Impact Level 5, TX-RAMP Level 2) are security authorizations. Neither the medical device page, the solution brief nor the platform terms states an intended purpose, excludes diagnosis or treatment, or says how automatic quarantine of a clinical device fits the customer's device risk management.

A certified AI management system, with no product evaluation published. The trust center lists ISO/IEC 42001:2023, the management system standard for artificial intelligence, among the company's certifications, and an AI section holding a platform AI disclosure FAQ and a data privacy and security statement for the AI assistant, both released on request.

The open face of the trust center gives neither the certificate's scope nor its date, so whether it covers the medical device product is unstated. Nothing is published on how device identification or anomaly detection performs across device classes, manufacturers or hospital network conditions, which is where a device security model is most likely to fail.

A warranty that the platform works as documented, with a correction remedy, and disclaimers over what it finds. The platform terms of 14 May 2026 warrant that current versions of the platform will perform materially in accordance with the documentation, with correction or replacement at no charge as the exclusive remedy, and that professional services will be re-performed or refunded.

They disclaim accuracy and any warranty that the solutions "will detect, prevent, or protect against all possible threats, vulnerabilities or other risks", and the indemnity covers intellectual property claims only. The warranty attaches to the software's conformance with its documentation, not to a classification, risk score or quarantine decision, so a device misidentified or wrongly cut off from the network has nothing standing behind it. The platform keeps a history of each asset's activity, which lets a customer trace what drove an alert. No accuracy, false positive or misclassification rate is published.

It has no clinical record surface and needs none; its integrations are with a hospital's security, IT and clinical engineering systems. The integrations page names Cisco ISE, Check Point, Splunk, ServiceNow, Jira, BMC, CrowdStrike, SentinelOne and Microsoft Defender among "hundreds of pre-built API integrations".

The medical device page says policies are enforced through the customer's firewalls, switches and network access control, and that "integrations with CMMS and ticketing systems automate the creation of work orders". ServiceNow, which completed its acquisition of Armis in April 2026, says the platform is integrated with the ServiceNow AI Platform. No CMMS product is named, no integration is documented as two way, and no named hospital is shown running a named integration.

Options and residency are stated; tenant isolation is not. The medical device page offers SaaS and on-premises deployment, and an on-premises subscription addendum dated 28 April 2025 is published. The subprocessor list dated 12 September 2026 lets the customer choose the United States, European Union, United Kingdom, Australia, Singapore or India for hosting, streaming, database and network services before the first term begins, and it shows what processing sits elsewhere: indicator of compromise classification by Google in the United States on IP addresses, MAC addresses and domain names, application security inference by Cerebras in the United States, and support from Armis entities in Israel, the United Kingdom, the Netherlands and the United States.

Regulated market customers are served from AWS GovCloud in the United States. Amazon Web Services hosts the service in a separate production environment. Nothing published says how one customer's data is kept apart from another's.

A published entry price and unit of charge, without tiers. The AWS Marketplace listing for the Armis Device Security Platform, sold by Armis, shows a "Centrix Standard minimum" of $3,250 per month on 1, 12, 24 or 36 month terms, billed on the assets and devices monitored, managed and unmanaged, including medical equipment, with private offers by email. The per asset rate, the price of the Medical Device Security product and what Centrix Standard includes are not published. The platform terms, a separate on-premises subscription addendum and the support terms are public.

Hospital settings are named with named deployments behind them. The medical device product is sold to hospitals and health systems for biomedical engineering and security teams, covering what the company lists as everything "from check-in kiosks to bedside monitors and surgical equipment", along with clinical assets and the network around them.

Deployments are named on the company's own pages: Main Line Health, Mater Misericordiae University Hospital, University Health Network and Burke Rehab Hospital, with a published case study on Princess Alexandra Hospital. The platform is also sold to manufacturing, energy, government, airline and education customers, and Medical Device Security is one of its products.