# 32 terms a reader meets, each with a sourced example

Glossary, read 10 Nov 2026

**The short answer.** The glossary defines 32 terms used in vulnerability remediation, from backlog and compensating control to risk acceptance and mean time to remediate. Each entry gives a short definition, a dated example with its source, and the read date of 10 Nov 2026. Terms are listed alphabetically.

- [Approval gate](#approval-gate)
- [Attack path](#attack-path)
- [Autonomy level](#autonomy-level)
- [Backlog](#backlog)
- [Base image](#base-image)
- [Choke point](#choke-point)
- [CMDB](#cmdb)
- [CVE](#cve)
- [Deduplication](#deduplication)
- [Exception](#exception)
- [Exploitability](#exploitability)
- [Fix group](#fix-group)
- [Fix verification](#fix-verification)
- [Human in the loop](#human-in-the-loop)
- [Infrastructure as code](#infrastructure-as-code)
- [KEV](#kev)
- [MCP server](#mcp-server)
- [Mean time to remediate](#mean-time-to-remediate)
- [Ownership resolution](#ownership-resolution)
- [Pull request remediation](#pull-request-remediation)
- [Reachability](#reachability)
- [Regression test](#regression-test)
- [Remediation](#remediation)
- [Remediation agent](#remediation-agent)
- [Ring deployment](#ring-deployment)
- [Rollback](#rollback)
- [Runtime context](#runtime-context)
- [SBOM](#sbom)
- [SLA](#sla)
- [Ticket bounce](#ticket-bounce)
- [Time to exploit](#time-to-exploit)
- [Zero-day](#zero-day)

## Approval gate

An approval gate is a point where an agent stops and waits for a person to approve before it acts. Gates can be set per action type, severity or asset sensitivity. The record of who approved what, with the evidence shown, is what an auditor reads later.

Example: Onit states a human approval gate on every agent action, with an audit trail and one-step reversal, for both remediations and mitigations (vendor's claim, read 10 Nov 2026). ([source](https://onit.security/platform/), read 10 Nov 2026)

## Attack path

An attack path is a chain of steps an attacker could take from an entry point to a valuable asset, passing through vulnerabilities, misconfigurations and permissions. Paths help choose which fix breaks the most routes, instead of fixing findings one at a time.

Example: Astelia describes attack path visualization to critical assets through network controls and open ports (vendor's claim, read 10 Nov 2026). ([source](https://www.astelia.io/platform), read 10 Nov 2026)

## Autonomy level

An autonomy level is the amount of independence an agent is given, from recommending only, through acting with approval, to acting alone inside set limits. Teams usually move up the levels over weeks as logged results show the agent's decisions are ones they would have made.

Example: Cogent describes staged autonomy: start with recommendations and review, then expand, with high-impact changes requiring human approval and an audit trail (vendor's claim, read 10 Nov 2026). ([source](https://www.cogent.com/ai/autonomy-levels), read 10 Nov 2026)

## Backlog

A backlog is the set of open vulnerability findings that a team has detected and not yet fixed, mitigated, accepted or dismissed. Its size in findings matters less than its size in analyst hours, which depends on minutes per finding and how many findings are duplicates of one fix.

Example: Averlon describes a customer example with 110,000 critical and high issues, 3,400 tickets and 260 pull requests; the customer is unnamed and the figures are the vendor's claim, read 10 Nov 2026. ([source](https://www.averlon.ai/blog/automating-remediation-application-infrastructure-security-issues), read 10 Nov 2026)

## Base image

A base image is the starting layer of a container image, such as an operating system layer. A vulnerability in it appears in every image built on it. Fixing the base image once can close hundreds of findings, which makes it a natural fix group.

Example: Averlon says it traces a base image vulnerability to the Dockerfile and routes it to the team that owns it (vendor's claim, read 10 Nov 2026). ([source](https://www.averlon.ai/blog/automating-remediation-application-infrastructure-security-issues), read 10 Nov 2026)

## Choke point

A choke point is a place where many attack paths pass through, such as one gateway, one account or one shared service. Applying a control or fix at a choke point can break many paths with one change, which makes it a cheap place to spend effort when no patch exists yet.

Example: Zafran describes applying a compensating control at the choke point of an attack chain, with a web application firewall rule as the example (vendor's claim, read 10 Nov 2026). ([source](https://www.zafran.io/resources/zafran-launches-attack-chain-killswitch-in-collaboration-with-google-threat-intelligence), read 10 Nov 2026)

## CMDB

A CMDB, or configuration management database, records assets and their relationships, often with an owner and criticality. It is the usual source for ownership, and the usual source of errors when it falls out of date after reorganizations or when cloud assets appear without a record.

Example: Seemplicity states that it identifies owners and criticality by syncing with CMDBs (vendor's claim, read 10 Nov 2026). ([source](https://seemplicity.ai/automated-vulnerability-remediation/), read 10 Nov 2026)

## CVE

A CVE is a public identifier for a specific vulnerability, in the form CVE-YEAR-NUMBER, assigned by a CVE Numbering Authority. The identifier is the key that links scanner findings, advisories, exploitation reports and patches. Not every exposure has one when it first appears.

Example: The Linux kernel CNA published CVE-2026-98432, a TCP use-after-free rated CVSS 9.8, on 15 Oct 2026, and Averlon published a response write-up on 2 Nov 2026. ([source](https://www.averlon.ai/blog/cve-2026-98432-linux-kernel-tcp-use-after-free), read 10 Nov 2026)

## Deduplication

Deduplication merges findings that describe the same flaw on the same asset when several scanners report it. It reduces the count without changing risk. It is the cheapest reduction in a backlog and should be checked before any claim of findings removed by analysis.

Example: Maze states that it pulls and deduplicates findings from the scanners a customer already runs (vendor's claim, read 10 Nov 2026). ([source](https://mazehq.com/platform/cloud), read 10 Nov 2026)

## Exception

An exception is a finding allowed to stay outside the normal SLA, with an approver, a reason and a review date. Exceptions should be few, written down and visible in the weekly meeting. A growing exception list shows a process problem that no scanner will fix.

Example: The PHP advisory for PHP2Shell lists an interim step, setting memory_limit to -1, and states it is not complete; an exception that relies on it should say so and carry a date to upgrade. ([source](https://github.com/php/php-src/security/advisories/GHSA-w8hw-w9fx-j79w), read 10 Nov 2026)

## Exploitability

Exploitability is whether a vulnerability can be used against a specific system, given its version, configuration, exposure and controls. It differs from severity, which describes the flaw in general. A critical flaw on an unreachable service can be low in exploitability for that environment.

Example: Maze defines a vulnerability as exploitable relative to the customer's own configuration, controls and context, not by CVSS or exploit-in-the-wild status (vendor's claim, read 10 Nov 2026). ([source](https://mazehq.com/platform/cloud), read 10 Nov 2026)

## Fix group

A fix group is a set of findings that one change closes, such as every CVE in a base image or a library version. Grouping by fix turns thousands of findings into a few hundred tasks, and it is how remediation platforms cut ticket volume.

Example: Zafran gives an example of 700 findings collapsing into a single OpenSSL upgrade ticket routed to the Platform team (vendor's claim, read 10 Nov 2026). ([source](https://www.zafran.io/), read 10 Nov 2026)

## Fix verification

Fix verification checks that a vulnerability is gone after a change, by rescan, version comparison or an exploit re-check. A closed ticket is not verification, because the ticket records intent and not the state of the system. NIST SP 800-40 Rev. 4 lists verifying installation among the steps of patch management.

Example: Qualys says it verifies that the vulnerability is fixed, rather than only the ticket closed, and rolls back automatically if a deployment causes issues (vendor's claim, read 10 Nov 2026). ([source](https://www.qualys.com/apps/agentic-ai), read 10 Nov 2026)

## Human in the loop

Human in the loop means a person reviews or approves an agent's step before it takes effect. It is a design choice, not a fault. The useful questions are where in the path the person sits and how many items they must review in a week.

Example: Astelia's example workflow leaves a drafted ticket and firewall rule in an approval queue waiting for a human click (vendor's claim, read 10 Nov 2026). ([source](https://www.astelia.io/post/closing-the-exposure-loop), read 10 Nov 2026)

## Infrastructure as code

Infrastructure as code defines cloud resources in files, such as Terraform, Pulumi or Helm charts. A misconfiguration fixed in the file stays fixed on the next deployment, which is why remediation tools open pull requests against these files instead of editing the live resource.

Example: Averlon states its fix actions cover container images, Terraform and Pulumi infrastructure as code, and Kubernetes Helm charts (vendor's claim, read 10 Nov 2026). ([source](https://www.averlon.ai/blog/automating-remediation-application-infrastructure-security-issues), read 10 Nov 2026)

## KEV

KEV is CISA's Known Exploited Vulnerabilities catalog: CVEs with reliable evidence of active exploitation. Under BOD 22-01, federal agencies must remediate listed items by the due date. Private teams often use it as the trigger for the emergency tier of an SLA.

Example: On 30 Oct 2026, CVE-2026-111357 and CVE-2026-118527, Ivanti Connect Secure flaws, were added to CISA's KEV catalog on evidence of active exploitation, as Rapid7 reported. ([source](https://www.rapid7.com/blog/post/etr-additional-ivanti-connect-secure-zero-days-exploited-in-the-wild-cve-2026-111357-cve-2026-118527/), read 10 Nov 2026)

## MCP server

An MCP server exposes a product's data and actions to AI assistants through the Model Context Protocol. For remediation tools it lets a customer connect its own language model and build workflows over the product's findings, subject to the permissions it grants.

Example: Tenable says its Hexa AI MCP server lets customers bring their own models and build custom workflows over Tenable tools (vendor's claim, read 10 Nov 2026). ([source](https://www.tenable.com/products/tenable-one/capabilities/hexa-ai), read 10 Nov 2026)

## Mean time to remediate

Mean time to remediate is the average time from a defined start, such as disclosure or detection, to a defined end, such as verified closure. It misleads when it counts only closed items, so a median, a 90th percentile and the age of open items should accompany it.

Example: Averlon states a customer's mean time to remediate fell from 45 days to 6 days; the customer is unnamed and the page gives no clock definition (vendor's claim, read 10 Nov 2026). ([source](https://www.averlon.ai/blog/automating-remediation-application-infrastructure-security-issues), read 10 Nov 2026)

## Ownership resolution

Ownership resolution is the step that decides which team can change the asset behind a finding. Sources include configuration databases, code owners, commit history, cloud tags and ticket history. A wrong answer produces a bounced ticket; an honest unknown can go to a default owner.

Example: Cogent's Discover page shows an agent resolving an asset owner from a ServiceNow CMDB and checking Jira ticket history (vendor's claim, read 10 Nov 2026). ([source](https://www.cogent.com/platform/discover), read 10 Nov 2026)

## Pull request remediation

Pull request remediation means an agent writes a code or configuration change and opens it as a pull request in the repository for review and merge. It fits dependency, container image and infrastructure-as-code findings, and keeps rollback inside the team's normal version control.

Example: Backline states that each remediation includes a plan and a ready-to-review pull request linked to the originating findings, and that existing unit tests run before delivery (vendor's claim, read 10 Nov 2026). ([source](https://backline.ai/product/), read 10 Nov 2026)

## Reachability

Reachability asks whether an attacker can get to the vulnerable code or service. Network reachability looks at routes and rules to the host. Code reachability looks at whether the application calls the vulnerable function. A finding can be present, and still not reachable.

Example: Onit describes a Reachability Agent that returns a determination per asset (reachable, restricted, internal, not reachable, unknown) with the path or the blocker as evidence (vendor's claim, read 10 Nov 2026). ([source](https://onit.security/blog/reachable-vulnerabilities/), read 10 Nov 2026)

## Regression test

A regression test checks that a fixed flaw stays fixed. When an agent writes a fix, a regression test written with it fails if a later change reopens the flaw. It is part of the evidence that a fix is more than a one-time edit.

Example: SiliconANGLE reports that Cogent's VR-1.1 model leaves regression tests that fail if a later code change reopens the flaw (third-party source, read 10 Nov 2026). ([source](https://siliconangle.com/2026/11/03/exclusive-cogent-securitys-vr-1-1-model-outscores-claude-code-and-codex-on-vulnerability-repair/), read 10 Nov 2026)

## Remediation

Remediation is the work that ends a vulnerability's exposure: applying a patch, changing a configuration, removing a component, or adding a control that blocks the path. It includes verifying that the change worked. Closing a ticket is not remediation unless the exposure is checked.

Example: Rapid7 reported on 2 Nov 2026 that Ivanti had released fixes for all three Ivanti Connect Secure vulnerabilities in an exploited chain, after the first fix on 13 Oct 2026 left the privilege escalation unpatched. ([source](https://www.rapid7.com/blog/post/etr-additional-ivanti-connect-secure-zero-days-exploited-in-the-wild-cve-2026-111357-cve-2026-118527/), read 10 Nov 2026)

## Remediation agent

A remediation agent is software, usually driven by a language model, that carries out one or more steps of fixing a finding, such as investigating, planning, writing a change or checking it. Vendors ship ready-made agents, and some let customers build their own.

Example: Backline names four vendor-built agents: Analyst, Planner, Coder and Verifier, with the Verifier validating every fix through testing (vendor's claim, read 10 Nov 2026). ([source](https://backline.ai/), read 10 Nov 2026)

## Ring deployment

Ring deployment rolls a change out in stages: a small test group first, then wider groups as health holds. It limits the damage of a bad patch. Platforms that deploy patches may pick the ring by a confidence rating for the patch.

Example: Qualys describes patch reliability scoring that sends high-confidence patches zero-touch, medium ones through staged rings and low ones to a manual change advisory board (vendor's claim, read 10 Nov 2026). ([source](https://www.qualys.com/apps/agentic-ai), read 10 Nov 2026)

## Rollback

Rollback returns a system to its state before a change. For an agent that deploys patches it is the safety net: a team should be able to trigger it, and the platform should trigger it on its own when health checks fail. Test it in a trial before relying on it.

Example: Tenable says Patch Management lets teams schedule, pause or roll back deployments under policy-based autonomous patching (vendor's claim, read 10 Nov 2026). ([source](https://www.tenable.com/products/patch-management), read 10 Nov 2026)

## Runtime context

Runtime context is information about what is running on an asset now: loaded libraries, open ports, active processes and effective configuration. It shows whether a vulnerable component is in use, which a static scan of installed packages cannot show. It is the main input for judging exploitability in a particular environment, and the least often shown to the customer.

Example: Maze states that it gathers runtime context without installing sensors (vendor's claim, read 10 Nov 2026). ([source](https://mazehq.com/platform/cloud), read 10 Nov 2026)

## SBOM

An SBOM, or software bill of materials, lists the components inside a piece of software. Matching a new advisory to SBOMs shows which products contain the flawed component before a scanner has a signature for it. The match depends on the SBOMs being current, so their upkeep is part of the cost.

Example: Astelia says its Hunter agent checks new and zero-day vulnerabilities against the SBOM before scanners detect them (vendor's claim, read 10 Nov 2026). ([source](https://www.astelia.io/use-cases/deploy-astelia-agents), read 10 Nov 2026)

## SLA

A remediation SLA is a time limit to fix or mitigate a finding, set by tier. It needs a stated clock start, such as disclosure or ticket creation, and a stated end, such as verified closure. Tiers by exploitation status hold up better than tiers by severity alone.

Example: CISA BOD 22-01 sets default remediation timelines for federal agencies of two weeks for known exploited vulnerabilities with CVEs assigned after 2021, and six months for older CVEs. ([source](https://web.archive.org/web/2024/https://www.cisa.gov/news-events/directives/bod-22-01-reducing-significant-risk-known-exploited-vulnerabilities), read 10 Nov 2026)

## Ticket bounce

A ticket bounce is a remediation ticket sent back or reassigned because it reached the wrong owner. The bounce rate, tickets reassigned divided by tickets opened, is a direct measure of ownership mapping and is worth tracking before and after any tool is added.

Example: Onit reports from its own field observation that roughly half of remediation tickets bounce because the owner is wrong on arrival; no method is published (vendor's claim, read 10 Nov 2026). ([source](https://onit.security/blog/zero-day-response/), read 10 Nov 2026)

## Time to exploit

Time to exploit is the interval between public disclosure of a vulnerability and the first observed exploitation. Averages hide a spread: some flaws are exploited before disclosure and others never. Shorter windows push teams to tier SLAs by exploitation status.

Example: Rapid7's Q3 2026 report puts the average at three days, with 71% of tracked exploitation in the first seven days, 80% within 14 and 90% within 30. ([source](https://www.rapid7.com/blog/post/tr-q3-2026-threat-update-attackers-exploiting-vulnerabilities-within-3-days/), read 10 Nov 2026)

## Zero-day

A zero-day is a vulnerability exploited, or exposed to exploitation, before a fix is available. Defenders have no patch and must rely on detection and controls. How fast a team can list its exposed assets decides how long it works blind.

Example: PHP2Shell leaked on 20 Oct 2026, drew exploit attempts overnight and was patched about 18 hours after the leak, according to Onit's 23 Oct 2026 write-up. ([source](https://onit.security/blog/zero-day-response/), read 10 Nov 2026)

## Related pages

- [31 questions answered](/faq/)
- [News](/news/)
- [Remediation Register: the data, downloadable](/datasets/)
- [How the register works, and what it leaves out](/method/)

## Sources

- https://onit.security/platform/ (read 10 Nov 2026)
- https://www.astelia.io/platform (read 10 Nov 2026)
- https://www.cogent.com/ai/autonomy-levels (read 10 Nov 2026)
- https://www.averlon.ai/blog/automating-remediation-application-infrastructure-security-issues (read 10 Nov 2026)
- https://www.zafran.io/resources/zafran-launches-attack-chain-killswitch-in-collaboration-with-google-threat-intelligence (read 10 Nov 2026)
- https://seemplicity.ai/automated-vulnerability-remediation/ (read 10 Nov 2026)
- https://www.averlon.ai/blog/cve-2026-98432-linux-kernel-tcp-use-after-free (read 10 Nov 2026)
- https://mazehq.com/platform/cloud (read 10 Nov 2026)
- https://github.com/php/php-src/security/advisories/GHSA-w8hw-w9fx-j79w (read 10 Nov 2026)
- https://www.zafran.io/ (read 10 Nov 2026)
- https://www.qualys.com/apps/agentic-ai (read 10 Nov 2026)
- https://www.astelia.io/post/closing-the-exposure-loop (read 10 Nov 2026)
- https://www.rapid7.com/blog/post/etr-additional-ivanti-connect-secure-zero-days-exploited-in-the-wild-cve-2026-111357-cve-2026-118527/ (read 10 Nov 2026)
- https://www.tenable.com/products/tenable-one/capabilities/hexa-ai (read 10 Nov 2026)
- https://www.cogent.com/platform/discover (read 10 Nov 2026)
- https://backline.ai/product/ (read 10 Nov 2026)
- https://onit.security/blog/reachable-vulnerabilities/ (read 10 Nov 2026)
- https://siliconangle.com/2026/11/03/exclusive-cogent-securitys-vr-1-1-model-outscores-claude-code-and-codex-on-vulnerability-repair/ (read 10 Nov 2026)
- https://backline.ai/ (read 10 Nov 2026)
- https://www.tenable.com/products/patch-management (read 10 Nov 2026)
- https://www.astelia.io/use-cases/deploy-astelia-agents (read 10 Nov 2026)
- https://web.archive.org/web/2024/https://www.cisa.gov/news-events/directives/bod-22-01-reducing-significant-risk-known-exploited-vulnerabilities (read 10 Nov 2026)
- https://onit.security/blog/zero-day-response/ (read 10 Nov 2026)
- https://www.rapid7.com/blog/post/tr-q3-2026-threat-update-attackers-exploiting-vulnerabilities-within-3-days/ (read 10 Nov 2026)

Last reviewed 10 Nov 2026. Canonical: https://automatedvulnerabilityremediation.com/glossary/
