nitsec Contact us Sign in
Agentic Penetration Testing

Every finding proven before you see it.

nitsec runs autonomous agents that test your applications the way a penetration tester would, then reproduces every finding before it reaches you and delivers a signed report with replayable proof. Continuously. On demand.

validated
Zero false positives by design
A finding that cannot be reproduced never reaches you.
graded
CVSS-grounded severity
Rated on what was demonstrated, not inflated.
provable
Replayable evidence
Every finding ships the traffic that proves it.
Shown, not told

This is what lands in your portal.

Not a scanner's alert stream — a validated set, each finding graded and carrying the exact traffic that proves it.

Engagement/Acme Bank/Finding Vulnerabilities
High Any customer can read another customer's account statements reproduced
CWE-639 · Authorization Bypass Through User-Controlled Key OWASP API1:2023 · Broken Object Level Authorization CVSS 8.1 confidence 0.95
Observed
The account_id is taken straight from the URL and the statements are returned without checking the account belongs to the signed-in customer.
Steps
Sign in as customer A → request GET /api/v2/accounts/50318/statements for an account A does not own → the response returns that account's statements in full.
Impact
Account ids are sequential, so any authenticated customer can walk the range and read every customer's balances, transactions and PII.
Remediation
Scope the query to the authenticated principal, or verify ownership of account_id before returning it.
Evidence3 exchanges · masked
POST/api/v2/session  sign in as customer A200setup
GET/api/v2/accounts/50318/statements  id A does not own200proof
GET/api/v2/accounts/50318/statements  unauthenticated401control
Authorization: Bearer •••••••• · Cookie: ••••••⧉ copy as curl

the findingClass and category by name, the reproduction in full, the impact and the fix — and the raw request/response sequence that proves it, replayable on demand with the negative control attached.

Acme Bank/Hunt/Vulnerabilities 5 findings
All 5 Critical 1 High 2 Low 2
CriticalBlind SQL injection in the transaction search filterCWE-89reproduced0.97
HighAny customer can read another customer's account statementsCWE-639reproduced0.95
HighSession cookie set without HttpOnly / SecureCWE-1004reproduced0.88
LowSignup form reveals whether an email is registeredCWE-204unvalidated—
LowLogout does not revoke the server-side sessionCWE-613reproduced0.82

the setThe count is findings, not noise. A low that could not be reproduced carries an unvalidated badge instead of being hidden or inflated — you see the engine's honesty, not just its hits.

Finding/Retest on-demand
REQUESTEDa developer pushed a fix and marked the issue resolved
Replay of the recorded exploit against the fixverdict in ~2 min
POST/api/v2/session  re-establish customer A200setup
GET/api/v2/accounts/50318/statements  id A does not own403exploit
✓ not reproduced — the fix holds. Finding moved to Fixed.status pushed to the linked issue

the loopClosing a finding is proof, not a checkbox. A retest re-runs the finding's own recorded exploit against the change — the 403 where there was a 200 is the evidence.

How a hunt works

Scope in. Signed report out. The loop stays closed.

A hunt is one autonomous engagement end to end: it maps the attack surface, tests through a recording proxy as your own principals, validates what it finds, and delivers.

01 scope

Authorize

Assets and rules of engagement, verified and locked.

02 hunt

Test

Agents probe each surface, signed in as real principals.

03 validate

Prove

Each candidate reproduced and graded. Unproven ones dropped.

04 report

Deliver

Risk-rated findings, evidence, remediation, coverage.

05 retest

Confirm

Ship a fix, get a verdict by replay — in minutes.

Why it's different

A report your team can act on the hour it lands.

Validated, not alerted

An independent pass reproduces every finding before delivery. You triage findings, never a scanner's backlog of maybes.

Severity you can defend

Impact is rated on what was actually demonstrated, on the CVSS scale — not pushed up a tier to look impressive.

Proof that replays

Each finding carries its real request and response traffic, masked for safety and replayable on demand. Verify it; hand it to a developer as a reproduction.

Fix it, then prove the fix

Request a retest and nitsec re-runs the finding's own exploit against your change and returns a verdict in minutes — not a new engagement.

Nothing out of scope

Scope and rules of engagement are enforced around every run. A host you did not authorize is never reached.

Continuous & on demand

Run on a schedule or on a deploy. A point-in-time pentest is obsolete the next commit; this keeps up.

Get started

See a hunt run on your own target.

Create an account and point nitsec at a staging target, or email us for a walkthrough and watch a finding reproduced in front of you. Walkthroughs take about thirty minutes.

app.nitsec.ai · sign in or create an account