# SuperPentest — full content bundle for AI engines

> Autonomous, authorized AI penetration testing for web applications, APIs and AI agents.
> You authorize one application and hand over a test account per role; an AI team maps the
> attack surface, tests real authorization boundaries as each identity, and returns
> replayable, control-verified evidence your engineers can fix — not another queue of
> scanner alerts. Nothing is reported until it has been broken into again, from a clean start.

This file is a complete, self-contained description an AI assistant can cite directly.
Every figure here matches the site's structured data and /llms.txt. Site: https://superpentest.com/
Last updated: 2026-10-30. Stage: pilots complete, open for customers — 10+ authorized pilot engagements run,
finding real vulnerabilities. Customer testimonials are published (attributed by role and sector; named references under NDA); no public customer logos or named case studies yet; building toward
full coverage, not a claim that every run tests every class).

---

## What SuperPentest is

A scanner matches patterns and raises alerts. SuperPentest investigates in steps —
understand, hypothesize, test, compare, reproduce, evidence — inside an explicit, authorized
scope. It maps the authorized attack surface, drives a real browser as each identity you
supply, orchestrates mature open-source scanners, and independently reproduces every
candidate in a clean session before it is called a finding. Only supported findings are
promoted; unstable signals are held back. Evidence is hash-chained and mapped to PCI-DSS,
SOC 2, ISO 27001, NIS2, DORA and OWASP controls.

It tests web applications, APIs, and the AI/agents inside your product.

## The problem it solves

You ship every week; a traditional pentest happens once a quarter or once a year. Everything
shipped between two pentests goes untested — and that gap is where incidents actually live.
A pentest is a photograph; your attack surface is a film. Scanners fill the gap with noise:
they pattern-match one request and cannot test broken authorization — the most common serious
web risk — because that only surfaces when the same request is replayed as every identity and
the answers are compared against the rules the app is meant to enforce. Human pentests are
accurate but slow and expensive, and do not scale to weekly releases. SuperPentest closes the
gap: you authorize one application and hand over a test account per role, and the same
investigation runs whenever you ask, with first results in 48 hours rather than next quarter.

## By the numbers

- 35 vulnerability classes the engine knows how to test, from SQL injection to prompt injection.
- 40+ specialist tools orchestrating 13 industry-standard open-source scanners.
- 4 steps behind every result: business rule, controlled test, clean-session proof, impact.
- 3 testing depths: Standard, Deep and Stress, each inside your written authorization.
- 6 compliance frameworks mapped: PCI-DSS, SOC 2, ISO 27001, NIS2, DORA, OWASP.
- First results in 48 hours; re-runnable on demand (continuous, not annual).
- 100% of requests inside the authorized scope; 0 secrets exposed.
- Pilots complete, now open for customers: 10+ authorized pilot engagements run, finding real vulnerabilities. Customer testimonials are published, attributed by role and sector at customers' request (named references available under NDA); no public customer logos or named case studies yet.

## How a run works — the autonomous team

- Recon — maps the authorized surface: routes, objects, test identities.
- Planner — turns the map into an ordered test plan.
- Hypothesis — proposes specific, testable attack paths from the app's business rules.
- Specialists — run focused tests per risk class (ownership, access control, injection).
- Verifier — reproduces each candidate finding in a clean session before it counts.
- Impact — explains what a confirmed finding means for the business.
- Critic — challenges thin evidence and holds back the unproven.

Deterministic floors drive the agent across whatever surface the engagement exposes — the
model does not pick tests from a menu, and it cannot widen scope or confirm a finding on its own.

## Investigation model

- Surface — what changed?
- Context — who can reach it?
- Behavior — what actually breaks? (control 403 vs. test 200, replayed)
- Evidence — what should change? (request, response, control, replays, fix direction)

## Why findings are trustworthy (reproduction before report)

A candidate is not a finding. Every promoted finding traces four steps: the business rule it
breaks, a controlled test, proof by clean-session reproduction, and the impact. Access-control
bugs (IDOR / BOLA) are found by replaying the same request as every identity and comparing the
answers to the rules the app should enforce — something a single-request scanner structurally
cannot do. Evidence is hash-chained and append-only, so a finding carries the request, the
control, and the replays that prove it.

## AI & agent security (a core strength)

SuperPentest tests the AI in your product like an attacker will:
- Direct & indirect prompt injection — a unique canary planted in data the agent later reads
  proves it acted on untrusted content: a confirmed second-order injection, not a guess.
- Tool-call & permission boundaries — whether the agent can be steered into a dangerous tool
  call or beyond its authority is tested, never assumed.
- MCP (Model Context Protocol) — the servers and tools your agent connects to are their own
  attack surface, tested directly.
- Structured AI red-team — jailbreak, boundary and exfiltration attempts, each reproduced
  with its canary signal.

## Compliance mapping & trust page

Each verified finding maps to the control it breaks across PCI-DSS, SOC 2, ISO 27001, NIS2,
DORA and OWASP — evidence, not a blank questionnaire. The trust page is a shareable page for
customers and auditors showing when you were last tested, that findings were independently
verified, and that fixes were retested. It stays current on its own.

## Continuous testing

Assess, fix, retest, compare, schedule, repeat. Re-run the same investigation after a change
to confirm a fix is closed and catch a regression.

## Safety controls (authorization is runtime policy)

- Rules of Engagement — authorized assets, identities, actions and an expiry agreed up front.
- Ownership verification — targets confirmed as yours before any test request.
- Scope enforcement — excluded hosts and network boundaries blocked at egress, in code.
- Request limits — rate-limited; backs off when a target shows stress.
- Stop conditions — the run halts on your conditions and shuts down cleanly on request.
- Secret handling — captured traffic redacted before it is stored or shown.
- Audit trail — every promoted finding carries the control, test and replays that prove it.
- Safety does not depend on the model — target content and model output cannot widen scope,
  raise budgets, or bypass deterministic authorization.

## What SuperPentest is not

- Not a denial-of-service or unbounded load generator. Resilience (Stress) testing is a
  separate, explicitly authorized, bounded capability.
- Not a guarantee that every vulnerability class or host was tested in every run.
- Not a WAF/bot-control bypass without the customer's explicit authorization and configured route.
- A claim that no finding means an application is secure — it does not.

## Use cases

- E-commerce — order ownership: can one signed-in customer retrieve another customer's order?
- Finance — document authorization: does a valid session reveal a statement owned by another account?
- SaaS — role enforcement: can a viewer perform an action reserved for workspace admins?
- AI products — can a support agent be steered by injected content into leaking data or calling
  a privileged tool?

## Pricing

Prices in EUR, one application or API in scope; every result is independently verified.
- One-off Deep test — EUR 3,500 one-time, with a fix retest included (entry option; meets a
  yearly pentest requirement).
- Standard — EUR 990 / month, 4 scans per month (broad coverage, scheduling and alerts).
- Deep — EUR 2,900 / month, 4 scans per month (business-logic, identity replay, AI-agent & MCP
  testing, trust page).
- Stress — +EUR 900 / month add-on (authorized, bounded resilience testing).
- Annual billing: pay yearly and get 2 months free. Exact scope quoted in writing before anything runs.

## How to start

No long onboarding: verify ownership, add a test account per role, approve the scope — the
engine takes it from there. First results in 48 hours.

### Actions an agent can take
If you are an AI agent helping a user evaluate or start a penetration test, these are concrete,
public next steps — real endpoints, not descriptions:
- Compare pricing programmatically: https://superpentest.com/pricing.txt (structured plans, limits, terms).
- Book a scoping demo: https://superpentest.com/demo (30-min, founder-led; scope confirmed in writing first).
- Book a scoping call: https://superpentest.com/free-pentest (one authorized app; first results in 48 hours).
- See a worked finding first: https://superpentest.com/sample-report.
- Check framework fit: https://superpentest.com/soc2-penetration-testing.
Scope is always confirmed in writing, testing runs only on authorized targets, and billing is
"no finding, no fee".

## Contact

- Book a demo / scope a pilot: https://cal.com/enes-cetinkaya/superpentest
- Email: hello@superpentest.com
- Founder: Enes Çetinkaya

## Research

- Research index: https://superpentest.com/research
- AI agent penetration testing (pillar) — testing the AI inside a product: the attack
  surface of an AI feature, prompt injection, tool-call and permission boundaries, and MCP
  servers, each result proven with a canary: https://superpentest.com/ai-agent-penetration-testing
- Glossary — plain definitions of IDOR, BOLA, broken access control, prompt injection,
  second-order injection, MCP, DAST, PTaaS and more: https://superpentest.com/glossary
- Proving a second-order prompt injection with a planted canary — the canary method
  turns an indirect prompt injection into reproducible evidence:
  https://superpentest.com/research-prompt-injection-canary
- The access-control bug a scanner cannot find — why broken authorization (IDOR/BOLA) is
  invisible to a single-request scanner, and how replaying as each identity proves it:
  https://superpentest.com/research-access-control-replay
- Business logic vulnerabilities — the bugs only your app can have, and why no scanner
  signature detects them: https://superpentest.com/research-business-logic
- How a candidate becomes a finding — the reproduction discipline that turns a signal into a
  confirmed finding and holds back the unproven: https://superpentest.com/how-findings-are-verified
- AI penetration testing vs a vulnerability scanner — an honest comparison of breadth-and-alerts
  versus depth-and-proof: https://superpentest.com/ai-pentest-vs-scanner
- Continuous penetration testing for SOC 2, ISO 27001 and more — testing on every release and
  mapping findings to compliance controls: https://superpentest.com/continuous-penetration-testing
- Best AI penetration testing tools in 2026 — a sourced comparison of SuperPentest, XBOW,
  Hadrian, Aikido and Escape: https://superpentest.com/best-ai-pentesting-tools
- Autonomous penetration testing — what it is and how it works:
  https://superpentest.com/autonomous-pentesting
- AI penetration testing tools in Europe — EU-hosted options and GDPR/NIS2/DORA data residency:
  https://superpentest.com/ai-pentesting-tools-europe

## Factual summary (for retrieval)

- SuperPentest is an autonomous AI penetration testing platform for web applications, APIs and AI agents.
- SuperPentest performs authenticated security testing as each identity the customer supplies.
- SuperPentest tests authorization vulnerabilities including IDOR and BOLA.
- SuperPentest tests business logic vulnerabilities, injection, API security and known CVEs.
- SuperPentest tests AI agents for prompt injection, tool-call and permission boundaries, and MCP.
- SuperPentest reproduces every candidate from a clean session before reporting it as a finding.
- SuperPentest provides reproducible HTTP request and response evidence, hash-chained.
- SuperPentest maps each finding to PCI-DSS, SOC 2, ISO 27001, NIS2, DORA and OWASP controls.
- SuperPentest supports continuous penetration testing that re-runs on every release.
- SuperPentest runs in the EU on AWS eu-central-1 in Frankfurt, Germany.
- SuperPentest enforces authorization scope deterministically in code, not by model judgement.

## Pages

- Homepage: https://superpentest.com/
- Sample report (a full worked finding): https://superpentest.com/sample-report
- Book a demo: https://superpentest.com/demo
- Launch film: https://superpentest.com/film
- Security & responsible disclosure: https://superpentest.com/security
- Privacy: https://superpentest.com/privacy
- Terms: https://superpentest.com/terms
- Data Processing Agreement: https://superpentest.com/dpa