Skip to content
How exposure is measured

Every layer of the device, and who can reach it.

The Secuno agent collects a device inventory every hour, with a full sweep every night, and a security heartbeat every 5 minutes. Then it answers the question inventory alone cannot: of everything on this device, what can an attacker actually reach?

app.secuno.nl/devices/acme-robotics
Device console showing a single endpoint's inventory, posture signals and agent heartbeat.
400K+
CVEs correlated
5 min
Heartbeat interval
1 hr
Inventory cycle, plus a nightly full sweep
0–4
Reachability tiers per device
Device visibility
Device visibility

Every layer of the endpoint, captured

The Secuno agent reads directly from the Windows registry, SMBIOS, WMI, and the Component Based Servicing stack, not via surface-level API queries. The result is a device picture no network scanner can produce.

OS identity & hardware baseline

OS name with accurate Windows 11 detection, build number, UBR revision, patch level, CPU architecture, IP and MAC. Hardware via SMBIOS: manufacturer, model, serial number, BIOS version, uptime, and last boot time.

Full software & patch inventory

Every installed application from all three Windows registry hives: 64-bit, 32-bit, and user-specific. Name, version, publisher, install date, architecture, and path. All KB articles with installation status read from the Component Based Servicing stack. Deduplicated by name, version, and publisher.

Security posture: AV, BitLocker, Firewall, Defender

Antivirus name, enabled state, and definition currency. BitLocker encryption method and conversion status per drive (XTS-AES-128/256). Windows Firewall state per network profile. Eight granular Windows Defender sub-protection states (realtime, behaviour, script scan, cloud protection, network IPS, and more) tracked individually and mapped to risk severity. Tamper protection state tracked; disabling it automatically opens a risk. Quick and full scan age collected and fed into the device compliance score.

Network adapters, open ports & processes

All IPv4 and IPv6 adapters with connection state and loopback detection. Open TCP and UDP ports with protocol, address, port number, PID, and process name. Full process list with PID, parent PID, and thread count via direct system enumeration.

Browser extensions: 6 browsers

Chrome, Edge, Brave, Vivaldi, Opera, and Firefox, read per user and profile. Extension ID, name including translated variants, version, and description. Risky or unknown extensions are visible immediately without any manual review.

Change detection across 8 dimensions

Every inventory cycle is compared against the prior measurement across software, KB patches, browser extensions, open ports, AV state, BitLocker, Firewall, and system identity. Every change is reported with old and new value and a precise timestamp.

Binary identity: SHA-256 on every launch

Process-creation and file-drop telemetry carry the SHA-256 and size of the image, so “the same thing is running on forty machines” is a fact about a file, not a guess about its name. A fleet-wide view shows where each binary runs, and every process row shows how common it is.

File integrity monitoring

The agent policy names the files and folders to watch. Every five minutes the agent reports what was created, changed or deleted, with the SHA-256 before and after, and it hashes every watched file at least once a day. Results are kept for 180 days.

Health score and end-of-support grading

A 0 to 100 health score heads each device, with a 90-day trend; a part with no data counts as not reported, never as failed. Operating system and software are graded against an end-of-support catalogue covering Windows, Windows Server, Office, SQL Server, Exchange and the .NET Framework, so a fully patched machine on a dead release still fails, and an unrecognised Windows build is not given the benefit of the doubt.

Attack-path exposure
Reachability, not port numbers

An open port is not an exposure until something can reach it

Asking whether a device listens on an alarming port number cannot tell a service bound to loopback apart from one hanging off the internet. Secuno asks three separable questions instead, and the answers multiply rather than pile up.

Tier 0: Not reachableThe service binds to loopback only. Nothing off this machine can talk to it.
Tier 1: Host-firewalledIt listens on a routable address, but the firewall profile in force has no inbound rule that lets anything in.
Tier 2: LAN reachableReachable from the local network segment the device sits on.
Tier 3: Untrusted networkThe device is on a network it does not control, such as a hotel or a client site, where the segment cannot be treated as friendly.
Tier 4: Internet facingReachable from outside the perimeter: the client's firewall allows it, a port forward points at it, or the agent has seen an internet peer connect to it.

Which way does the attack come in?

The attack path is derived from the CVSS vector, not the score. A network RCE against a listening service and a browser bug that needs a user to click are both nominally “network” and are nothing alike. A CVE that carries no vector string is recorded as unknown rather than assumed.

Can an attacker actually reach it?

A reachability tier from 0 to 4 is built from socket bind addresses, firewall profile state, inbound-allow rules, adapter addressing, the inbound peers the agent has actually seen and the client's perimeter firewall. Tier confidence is tracked separately, so a device that has not checked in for six weeks is not scored as though its port list were current.

Do the controls block that path?

A control only counts against the vector it actually blocks. Disk encryption stops a stolen laptop and does nothing for a network RCE. Exposure and controls are multiplicative rather than additive, so a fully hardened host can no longer outrank a bare one.

Which device drives the score?

When one vulnerability affects many devices, the worst affected device sets the score. Device groups add what no agent can measure, such as “not reachable over the network”, “this is the DMZ” or “business critical”, as bounded score adjustments listed on the risk apart from the measurements, and a multi-device risk is assessed against the device its groups call most exposed.

A documented perimeter, not an aspiration

Inbound filtering, segmentation and egress control are things an endpoint agent structurally cannot observe. Each client documents its firewall in a thirty-question questionnaire an operator can actually answer and someone else can audit, or has it read from the firewall itself, and the scoring axes are derived from the answers, each shown with the sentence it came from, rather than asked outright. The result caps the tier for every device behind it, and a record nobody has checked against the running configuration is discounted where that changes the answer.

Read from the firewall itself

With the firewall module switched on for a client, a read-only connection to FortiGate, Meraki MX, SonicWall or Sophos Firewall answers what the questionnaire asks; a credential that can write is refused. Measured answers fill and lock the questionnaire, and a port forward to a listening port makes that device internet facing. A FortiGate read also covers address objects, local-in policies and whether IPS, DNS filtering, web filtering and antivirus are on; Meraki administrators are judged on the privileges they actually hold. Configuration risks open for exposed management, admins without MFA, vulnerable or unsupported firmware, expired licences, risky forwards, unused policies, VPN without MFA, weak IPsec and logs kept only on the box. IPS, malware, VPN and admin events are read every five minutes and open or join cases; ordinary traffic is never stored. Configuration drift is recorded with who made the change.

What the firewall knows that the agent cannot

The firewall's own device list is compared with the managed devices, and computers seen regularly without the Secuno agent raise one grouped risk per client; guest networks, phones and randomised hardware addresses are never kept. A device's egress, segmentation and IPS credit comes from the policy that actually covers its network rather than the site questionnaire. Once a day Secuno completes a TCP handshake against the site's public addresses, and a port open from outside that the configuration does not explain becomes a risk, critical for RDP, SMB and Telnet. Firmware flaws are read from vendors' own advisories, so they can reach a risk before NVD has mapped them, and a firewall C2 or malware hit the agent also saw counts as confirmed. Device discovery and the outside-in check are off until switched on per connection.

Perimeter response, only with separate consent

Reading stays read-only. Where a client gives separate consent, Secuno can maintain one blocklist group and one quarantine group on a FortiGate or Meraki. It never touches the client's own policies, and the first push waits for a person to approve a dry run.

Measured, not only declared

Agents report who connects in to each listening service, classed as on-subnet, private or internet, without probing anything. An observation can only raise a tier: one internet peer makes the device internet facing whatever was declared, and raises a single high “declared perimeter contradicted” risk on the firewall record, which closes only once the record is corrected and no internet peer is seen. The exposure card shows what was observed (the listener, its peer counts and when it was last seen) above what was inferred, and a connection from another managed device of the same client becomes a measured path. On Linux the agent reports which host firewall is in use, and says so plainly when none is filtering inbound traffic.

How far could it spread?

Per client, Secuno shows which workstations accept SMB, RPC, RDP, WinRM or SSH from the local network, and how many of those paths were actually seen in use. It is raised as one lateral surface risk per client: the map an attacker would use after the first foothold, drawn before they need it.

The evidence, not just the number

Each device stores its tier and the assessment behind it, so the list sorts and filters on exposure directly, and a risk can still show the evidence its score was built from after the device has since changed. A device with no location is offered the one its gateway and network neighbours share, so it is scored against the right firewall; a person confirms the assignment.

Web exposure
Outside-in, per client

What a client's websites show the internet

The agent sees a device from the inside. Web Exposure checks each client's published websites from the outside, within a deliberately narrow boundary, and turns what it finds into evidenced findings, a grade from A to F and scored risks in the same register.

AI Exposure
Opt-in module, per client

Which AI is in use, and what it can reach

AI Exposure finds the AI tools on a client's devices and in its Microsoft 365 tenant from data Secuno already collects, and scores what is exposed the same way as any other finding. It is an opt-in module, switched on per client at no extra cost. It checks how Copilot is set up, not what people ask it: Secuno records no Copilot activity.

51 AI tools, found without a new agent

A catalogue of local model servers (Ollama, LM Studio, vLLM, Open WebUI and others), desktop clients, coding assistants, browser extensions, AI services and meeting bots, matched after every inventory against installed software, running processes, listening ports and browser extensions, plus DNS where monitoring is on.

Model servers graded by who can reach them

A local model server is judged on the same reachability tiers as everything else: high when it answers beyond the device itself, critical on an internet-facing host or behind a port forward, lower when it requires accounts. A signature port without a matching process is flagged as a possible AI server.

AI apps users approved in Entra

Meeting bots and AI services that people consented to in Microsoft 365 are judged by the access they hold. A meeting bot with calendar, meeting or mail access is high; one with file or site access is medium; the rest is inventory.

Copilot posture, six checks

A new AI control section checks for a privileged role on a Copilot-licensed account, public Microsoft 365 groups, no sensitivity labels, an agent deployed to the whole organisation, agent requests left waiting and concealed report names. A Copilot seat raises an existing missing-MFA finding one step instead of duplicating it.

Readiness before the first seat

On a tenant without Copilot seats the same checks run as readiness results: no risks, a checklist and a PDF of what to fix before assigning licences. What app-only access cannot read is listed as a manual check, never as a pass. A Copilot-only tenant needs no agent.

Across the portfolio, with sanctions

AI columns on the exposure portfolio show open critical and high AI risks, Copilot seats used, assigned and bought, and exposed model servers, with no composite score. A source that is switched off reads off, never 0. A tool the client sanctions, with a reason and an expiry, leaves the shadow-AI risks but never hides a misconfiguration.

From exposure to action
Beyond visibility

Every exposure becomes a scored, owner-assigned risk

Every data point collected here flows automatically into CVE correlation and risk scoring. The full correlation engine, scoring, and remediation workflow live on the Vulnerability Management page; the AI and infrastructure model lives on the Security page.

Fleet operations
Fleet operations

Operational control over every agent

Deploy, quarantine, revoke, and monitor agents across your entire client portfolio from a single interface. No endpoint access required for fleet management.

Enrollment tokens

Create tokens with label, expiry date, and maximum use count. Shown once in the UI and never retrievable again. Only the hashed value is stored server-side. Prevents unauthorised agent registration without any agent-side configuration.

Quarantine & revocation

Place individual devices in quarantine, lift it, or permanently revoke them from the device overview. Direct operational control over your agent fleet without touching the endpoint itself.

Offline queue, no data loss

Up to 1,000 events are stored locally on the device when the agent loses connectivity. Sent in sequence upon reconnection. Inventory continuity is maintained through temporary network outages without operator intervention.

Configurable offline alerting

Agents that stop reporting trigger a notification after a configurable threshold: default 20 minutes, adjustable per partner environment from Partner Settings. Fires once per silence period, not on every check cycle.

See the agent in action

Full device visibility across your entire client portfolio

We walk you through agent deployment, CVE correlation, and risk workflow live, in a demo environment matched to your portfolio size.