PhantomDragon AI

The scanning engine behind every test we run.

PhantomDragon AI is the scanning engine our engineer runs at the start of every penetration test: 50 modules across a 17-phase pipeline, running 75 or more scanners. Its detection is deterministic, built from payloads, probes, an offline CVE database, and scanners for JavaScript, authentication, protocols and web application firewalls. It does the coverage work in a few hours; the engineer then confirms every finding by hand before it reaches a report, and tests what a scanner cannot.

It is not sold on its own: it runs inside our security scan and our penetration test. Not sure you need either yet? Start with the free scan of your website.

Run the free scanHow a pentest works
50
Scanning modules
17
Phase pipeline
75+
Scanners it runs
Detection

How it finds things

Six things to know about the engine, including what it leaves to a person. The full method it sits inside is published.

  1. Deterministic detection

    Findings come from payloads, probes and the target's own responses, not from a model's opinion.

  2. An offline CVE database

    The software it fingerprints is checked against a vulnerability database held offline, alongside the engine.

  3. Specialist scanners

    Scanners for JavaScript, authentication, protocols and web application firewalls, among the 75 or more the engine runs.

  1. Plain about its limits

    Some weaknesses only show with two identities: one tenant reaching another's data, or one role using another's functions. From one role the engine reports those as not tested rather than claiming them. The engineer tests them by hand, with two roles.

  2. AI only when it is switched on

    An optional AI layer can help with triage, ideas for getting past a web application firewall, and hypotheses worth testing. It is off by default, each part has a deterministic fallback, and the engine never downloads a model on its own.

  3. A person confirms every finding

    Nothing reaches a report on the engine's word. The engineer reproduces each finding by hand, with the request and the response, or drops it.

On our planted benchmark, a target with six classes of weakness planted in it, the engine found all six and raised no false-positive candidates. That measures the engine against flaws we planted, not against your application, which is why an engineer still confirms every finding.

Execution

Where it sits in a test

1ScopeThe engineer agrees the scope and the written authorisation before anything runs.
2SweepThe engine runs its 17 phases against the scope while the engineer watches. It takes a few hours.
3ConfirmEvery lead is reproduced by hand or dropped. What one role cannot show is tested by hand with two.
4ReportA PDF report, with JSON and SARIF files for your developers. Every finding says whether the engine found it and an engineer verified it, or the engineer found it.

What the report looks like, page by page, is in our example report. How automated and manual testing compare is in AI vs manual penetration testing.

Questions

Questions about the engine

What is PhantomDragon AI?

PhantomDragon AI is the scanning engine our engineer runs at the start of every penetration test: 50 modules across a 17-phase pipeline, running 75 or more scanners. Its detection is deterministic, built from payloads, probes, an offline CVE database, and scanners for JavaScript, authentication, protocols and web application firewalls. It does the coverage work in a few hours; the engineer then confirms every finding by hand before it reaches a report, and tests what a scanner cannot.

Does PhantomDragon AI use AI to find vulnerabilities?

Not by default. Detection is deterministic: payloads, probes, an offline CVE database and specialist scanners. An optional AI layer, for triage, ideas for getting past a web application firewall and hypotheses worth testing, is off by default and switched on per run; each part has a deterministic fallback, and the engine never downloads a model on its own.

Is it fully autonomous?

No. The 17-phase pipeline automates scanning end to end, but a Ghost Protocol engineer scopes every engagement, drives the run, and verifies every finding before it reaches your report.

Can I buy PhantomDragon AI on its own?

No. It runs inside our security scan and our penetration tests, driven by an engineer who verifies every finding.

How accurate is it?

On our planted benchmark, a target with six classes of weakness planted in it, the engine found all six and raised no false-positive candidates. That measures the engine against flaws we planted, not against your application, which is why an engineer still confirms every finding.

What can it not test?

Anything that needs two identities to see, such as one tenant reaching another's data, or one role using another's functions (BOLA and BFLA). From one role the engine reports those as not tested, and the engineer tests them by hand with two roles. Our engagements also leave out people, premises and internal networks.

Book a penetration test

Email support@ghosts.lk or message +94 71 055 5055 on WhatsApp. We reply within 24 hours. Working hours are Monday to Friday, 09:00 to 18:00 Colombo time (UTC+5:30).