Two Assets, the Same Critical Vulnerability. An Attacker Can Reach Only One.

Scanners give a vulnerability the same score on every server, whether it's a hop from the internet or behind three firewalls. Onit's Reachability Agent shows which servers an attacker can actually reach.

Onit Security | Nov 6, 2026
A panther with glowing orange geometric markings crouches on the edge of a cracked rock ledge against a dark purple background.

Key takeaways

  • A vulnerability's score is the same on every asset. Whether an attacker can get to that asset depends on what sits between the attacker and the vulnerable code.
  • "Internet-facing" is rarely a yes or no. A server can be open to anyone, open only to named sources, reachable from inside, not reachable, or unknown, and each calls for a different decision.
  • October's exploited edge flaws made reachability the first question. Mandiant's advice on BeyondTrust: treat any appliance that was reachable from the internet on a vulnerable version as possibly compromised until reviewed.
  • Onit's Reachability Agent gives every asset a verdict without changing anything, with the path, or what blocks it, as evidence. 

Two servers run the same version of the same software and carry the same critical score. One answers on port 443 through the edge load balancer. The other sits in a segment no outside path reaches. Your scanner lists them as equals.

Same score, different risk

A severity score describes the flaw, not where it sits. It can't tell you that a server tagged "internal" faces the internet through one load balancer rule nobody remembers adding, or that a medium-severity flaw has an open path in.

In October, the advice on exploited appliances came down to reachability

BeyondTrust disclosed CVE-2026-102764 in Remote Support on October 15. A China-nexus espionage group had exploited it as a zero-day since late August, and Shadowserver counted more than 1,400 instances exposed online. Mandiant's advice went past patching: "Any appliance that was reachable from the internet on a vulnerable version during the exploitation window should be treated as potentially compromised until it has been reviewed."

Exploiting a vulnerability is now the most common way into a breach, at 31% of initial access in the 2026 Verizon DBIR. And Mandiant's advice can't be followed from the advisory alone. Whether your appliance is reachable today depends on your own controls and infrastructure.

What makes a vulnerability reachable?

Reachable vulnerabilities are the ones an attacker can actually get to from somewhere they can stand. The internet is one starting point. User networks, partner links and other segments are others, because an attacker who gets in at the edge moves east-west from there.

VerdictWhat it meansWhat it usually leads to
Reachable from the internetOpen to any sourceFix now
RestrictedReachable from the internet, but only from named addresses, such as a partner'sA short deadline, or confirm the allowlist
Reachable internallyFrom a user network, a partner link or another segmentThis week's window
Not reachableNo path from any attacker starting pointAccept for now, with the evidence on record
UnknownThe configuration needed to decide is missingTreat as reachable

A yes-or-no tag can't express restricted, internal-only or unknown. Working out which verdict applies, for every asset, is what reachability analysis does.

Why does reachability cut so much of the noise?

Most vulnerable assets can't be reached by an attacker at all. Scanners find the vulnerable version on every asset, but not what stands between an attacker and each one, so every finding lands in the queue looking equally urgent. Check reachability and most of that noise drops away. What's left is the short list an attacker can actually get to.

That matters because no team fixes everything. The 2026 Verizon DBIR found only 26% of known-exploited vulnerabilities fully remediated, down from 38%. Every hour spent on an asset nobody can reach is an hour the reachable ones don't get.

The gap shows up on the worst days. A Fortune 500 enterprise we work with got no findings from any of its three security tools on the morning a flaw in one of its products entered CISA's Known Exploited Vulnerabilities catalog. Onit found two devices running that product. One could be reached from the internet, and none of the three tools covered the other.

How does the Reachability Agent work?

The Reachability Agent reads the rules and configurations you already have, joins them to every asset, and returns a verdict with its evidence for every exposure.

  1. It reads your environment and traces the paths, read-only. It reads your environment as deployed, links what it finds to the vulnerable services, and traces paths from the internet, user networks, partner links and other segments.
  2. It returns a verdict with its evidence. A reachable verdict comes with the path, hop by hop. A not reachable verdict comes with what blocks it. An unknown verdict names the device or configuration that would settle it.
  3. It checks again when anything changes. Full collection happens in minutes, and verdicts are recomputed whenever new configurations arrive, with anything that just became reachable flagged.
  4. It feeds every exposure's assessment. Verdicts go into the overall assessment of each exposure in the Decision Center, alongside the other agents' answers.

What it doesn't do. It doesn't guess at your rules: parsing and path tracing are exact logic, not a model's reading of a config. It never changes a rule, installs nothing on your servers and sends no traffic to test a path. Customer data never trains models, and Onit holds SOC 2 Type II, ISO 27001 and ISO 42001.

From verdicts to decisions

In Decision-Based Exposure Management, a reachability verdict is the first of four answers behind every decision. The unit of work is the decision, not the ticket: a person makes one judgment about a whole class of exposures, and agents carry it to every matching exposure, including ones that haven't appeared yet.

The four answers, in order:

  1. Reachable: can an attacker actually get to the vulnerable code? The Reachability Agent answers it.
  2. Exploitable here: does the attack work on this server, or does a control already block it? The Exploitability Agent checks, read-only, through the EDR you already run.
  3. Business context: how critical is the server, and what breaks if it stops?
  4. Ownership: who signs off on the fix, drawn from live context such as CMDB records and ticket history rather than a list that went stale.

Take a critical flaw on 40 servers. Instead of 40 tickets, the team gets three decisions:

  • 6 reachable from the internet, attack conditions present, serving customers: the web platform team patches today.
  • 12 reachable from the user network, running internal finance apps: their owners patch in this week's change window.
  • 22 in an isolated build segment with no path in: accepted for now, with proof and an expiry date.

Microsoft pointed the same way in October, grouping its record Patch Tuesday into "risk cohorts" around shared questions such as which remote access roles are "deployed and exposed," and telling customers to "answer the shared environment question once and document the result."

A decision goes one step further: it works as an operating rule. Next month, a new build server with the same flaw inherits the decision made for its segment. If a firewall change opens a path to one of the 22, that decision reopens with the rule that changed. And the reasoning stays on record after the analyst who made the call moves on.

What this means for security teams

Ask for the path, not the tag. "Reachable from the internet through the edge load balancer on port 443" gives IT something to check. "Internet-facing: yes" doesn't.

Check ten "internal" servers. Pick ten servers your CMDB marks as internal and check, rule by rule, whether any of them can be reached from the internet.

Treat unknown as reachable until the missing configuration is in.

Then start with your mediums, and ask which of them an attacker can reach from the internet today.

Frequently asked questions

An attacker can get to it from somewhere they can stand: the internet, a user network, a partner link or another segment. Reachability is separate from severity. A critical flaw nobody can reach can wait. A medium one with an open way in can’t.

Trace the path instead of trusting the tag: the edge firewall and NAT rules, load balancer listeners, WAF policy and cloud security groups in front of the server, down to the port the vulnerable service listens on. Then check which sources the path allows. A server open to the whole internet and one open only to named partner addresses are different decisions.

No. It means no network path exists today from the attacker starting points that were checked. The exposure should close under a recorded decision, with the blocking rules as evidence, and reopen if a rule change opens a path. Paths through stolen credentials are a separate question.

Mandiant said any BeyondTrust Remote Support appliance that was reachable from the internet on a vulnerable version during the exploitation window should be treated as potentially compromised until it has been reviewed. The flaw was exploited as a zero-day from late August 2026 and disclosed on October 15, 2026.

An agent in Onit’s Decision-Based Exposure Management platform. It reads your environment read-only, traces paths from the internet, user networks and partner links to every asset, and returns a verdict with its evidence: reachable from the internet, restricted to named sources, reachable internally, not reachable, or unknown. Verdicts feed the overall assessment of each exposure in the Decision Center.

Get a Demo