Glossary / AIpento

Security testing, in plain words.

The terms you will meet when you buy, run or read a security test, from SAST to SOC 2. Each in a sentence or two, with where AIpento fits.

AI penetration test
A penetration test carried out by AI agents: they map the application, plan, run tests and chain findings. AIpento keeps only findings it can reproduce, each with the request and response that proved it.
Attack surface
Everything about an application that someone outside can reach or influence: pages, API endpoints, inputs, sign-in points, file uploads, and anything that talks to an AI model.
BOLA
Broken object level authorization. The API name for an IDOR.
Broken access control
Any case where a user can do or see something they should not: another customer's data, an admin action, a feature they did not pay for.
CORS misconfiguration
A setting that lets other websites call your API from a visitor's browser. Allowing every origin together with credentials is the dangerous case.
CVSS
Common Vulnerability Scoring System. A standard way to score how severe a vulnerability is, from 0 to 10.
CWE
Common Weakness Enumeration. A catalogue of software weakness types, each with a number, used to label findings consistently.
DAST
Dynamic application security testing. Sends requests to a running application and studies the responses, as an outsider would.
Detector
The one named rule or AI check that found a particular problem - a SQL-injection check, a leaked-secret check, a broken-access-control check. Every finding on the Findings screen carries the detector that fired, so matching findings group together.
Evidence Pack
The record an auditor needs: what was tested, what was found, the proof for each finding, and what was fixed.
Exploit chain
Several weaknesses, each minor on its own, used together to reach something serious.
False positive
A reported finding that is not a real problem. Every one costs review time and erodes trust in the rest of the report.
IDOR
Insecure direct object reference. The app uses an identifier from the request to fetch a record without checking that it belongs to the person asking.
Infrastructure as code (IaC) scanning
Checking the files that build and run your app, such as Dockerfiles, CI workflows, Terraform and Kubernetes manifests, for unsafe settings.
Injection
Untrusted input being treated as part of a command or query. SQL injection is the best-known kind.
MCP server
A Model Context Protocol server: a way of giving AI agents tools they can call. Their code gets its own checks, because a careless tool hands an agent too much power.
MTTR (mean time to remediate)
The average time from a finding opening to it being fixed and accepted. Shown on the Findings dashboard; a shorter MTTR means problems are not sitting open.
OWASP API Security Top 10
The same idea for APIs. The 2023 edition puts broken object level authorization first.
OWASP Top 10
A list of the ten most critical web application security risks, published by the OWASP Foundation. The 2021 edition puts broken access control first.
OWASP Top 10 for LLM applications
OWASP's list of the main risks in software that uses large language models, from prompt injection to excessive agency. AIpento's live tests include checks mapped to it.
Penetration test (pentest)
An authorised, simulated attack on a system to find weaknesses before someone else does. It ends with a report of what was found and how it was proven.
Privilege escalation
Gaining more rights than you were given: a member acting as an admin, or one customer acting inside another's account.
Prompt injection
Text that tricks an AI feature into ignoring its instructions: revealing data, taking actions it should not, or saying what an attacker wants.
Proof of exploit
Evidence that a finding is real: here, the exact request sent and the response that came back.
Release gate
A check that a release must pass before it ships. AIpento's answer is one of two words: Ship, or Fix first.
Remediation
Fixing what was found. AutoFix writes the patch; it is accepted only when the test suite passes three times from three clean checkouts.
Retest
Running the test again after a fix to confirm the problem is gone.
SAST
Static application security testing. Reads source code without running it, and points to the file and line of a risky pattern.
SCA
Software composition analysis. Checks the third-party packages you depend on against databases of known vulnerabilities. AIpento checks npm and PyPI packages against the public OSV database.
Scope
The written list of what a test may touch. In AIpento the scope is enforced in code, and everything done inside it goes into an audit trail.
Secrets detection
Finding keys, passwords and tokens that were written into code or configuration, including in old commits that are still in the history.
SOC 2 / ISO 27001
Security standards many business customers ask their suppliers to meet. Auditors for both often expect regular penetration testing. AIpento maps its evidence to their controls; it does not certify anyone, and holds neither certification itself.
SSRF
Server-side request forgery. An attacker makes your server send a request on their behalf, often to internal systems that are not meant to be reachable.
Staging environment
A copy of an application that is not serving real customers, used for testing. AIpento's live tests run against staging, never production.
Verified three times
AIpento's bar for trusting an AutoFix patch: the test suite must pass on three separate, clean checkouts before the fix is accepted. One green run could be luck; three in a row is not.
XSS
Cross-site scripting. Text from one user is written into a page as HTML, so a script inside it runs in another user's browser.