Method

AI-agent penetration testing method: The pipeline

  1. 01/ 04

    Custom scope

    Together, in black and white, we define exactly what we're allowed to attack: your website, your code, your IP addresses, your subdomains. Everything that matters is covered, nothing goes beyond what was agreed.

  2. 02/ 04

    In-depth recon

    We map everything visible from your target: site pages and user flows, code, exposed services, domain names, public information. We rebuild the entire application before launching a single attack.

  3. 03/ 04

    40 agents in parallel

    40 specialized agents attack at the same time, each on its own turf: access and identity, infrastructure, injections, business logic flaws, exploit chains. Above all, they collaborate in real time: what one agent uncovers, another uses to go further. Each one is worth a senior consultant.

  4. 04/ 04

    Two reports

    Two reports: a technical one for your teams (vulnerabilities, proof, fixes) and a compliance one for your leadership: risk translated into quantified regulatory exposure (GDPR, ISO 27001, NIS2). Board-ready, not an unreadable spreadsheet.

Safety

How we operate safely

We really do attack your systems, but never without guardrails. Here are the safeguards that protect your production environment on every engagement.

  1. 01 · Authorization

    Written authorization first

    The attack only starts once you've signed off in writing (rules of engagement + DPA), on a scope defined with you.

  2. 02 · Environment

    Controlled environment

    Reversible actions, no denial of service, can be stopped at any time, no data tampering. Non-destructive attacks.

  3. 03 · Control

    You stay in control

    You decide the timing and the extent of every engagement.

  4. 04 · Reports

    Two reports delivered

    A compliance report your leadership (CEO/CTO) can actually read, and a technical report for your teams (CISO).

Our approach

What makes the difference

  1. 01 · Depth

    In-depth detection

    Beyond the usual CVEs and misconfigurations, our agents find business logic flaws, broken access control, mass assignment and exploit chains. Exactly the kind of vulnerabilities behind most breaches today.

  2. 02 · Reasoning

    Offensive reasoning

    Each agent specializes in one specific surface and thinks like a determined attacker: chaining findings, exploiting forgotten corners, confirming real exploitability instead of churning out false positives.

  3. 03 · On demand

    On-demand pentests

    Launch a pentest the same day, with no scoping phase and no paperwork delays. Your security cycle follows your release cycles and sprints, not the other way around.

  4. 04 · Anticipation

    Get ahead, don't play catch-up

    A critical vulnerability that goes undetected turns into a data breach, a regulatory fine or reputational damage, not to mention the cost that comes with it. Continuous pentesting means finding your vulnerabilities before an attacker does.

Trust

Our commitments

Independence

The people who find your vulnerabilities aren't the ones billing you to fix them. Remediation is a separate contract: that's what guarantees an honest assessment.

Honesty

No test can guarantee zero vulnerabilities; we attest to a scope, on a given date. No false promises.

Retest everything, every time

Every pentest retests your entire scope, not just what changed: a single modified feature can affect all the others.

Case studies

Sample findings

takeover · 4.8s
$ # step 1: create an account
POST /auth/signup → 201 Created
$ # step 2: escalate the role
PATCH /users/me {"role":"admin"} → 200 OK
$ # step 3: admin access
GET /admin/dashboard→ 200 ✓ ADMIN
Critical · Privilege Escalation
Critical · Privilege Escalation

Admin takeover in 3 HTTP requests

Public signup combined with overly permissive RLS UPDATE policies. An anonymous visitor becomes CRM admin in seconds. No password or invitation required. Root cause plus 14 derived findings detected by the identity agent.

Request a similar pentest
SELECT * FROM clients212 / 212 rows
Full PII
Exact GPS
408 orders
Critical · Data Exposure
Critical · Data Exposure

Massive GDPR breach: 212 B2B customers

No user_id filter in the business RLS SELECT policies. 212 customers with full PII, GPS coordinates, 408 orders and 56 VP contacts exposed to any authenticated user. A severe GDPR violation, with a direct CNIL fine (France's data protection authority) on the table.

Request a similar pentest
From:direction@showroomaugust.fr✓ SPF✓ DKIM
To: 212 B2B customers
Subject: Urgent Q2 payment validation
Hello, following an incident with our payment provider, please re-confirm your bank details by clicking the secure link below before 6 p.m. today.
Re-confirm my payment
Sent via Edge Function /send-email · DMARC p=none
Critical · Email Abuse
Critical · Email Abuse

Phishing from legitimate infrastructure

The email Edge Functions are reachable by any authenticated user, with no server-side admin role check. The target domain's legitimate SPF and DKIM can be abused. Convincing fraud campaigns, helped by DMARC p=none.

Request a similar pentest
User field · prospects.site_web
javascript:fetch('/exfil', {method:'POST',body:document.cookie})
<a href=payload>Visit website</a>
Cookie stolen when the sales rep clicks
High · Client-side Attack
High · Client-side Attack

Stored XSS via a free-text user field

javascript: payload rendered without validation in a user-supplied URL. RLS INSERT/UPDATE left open. Cookie stolen in the context of a sales rep who clicks the record. A CSP with unsafe-inline weakens the protection.

Request a similar pentest