Methodology

How we test: people decide, the platform executes, evidence settles it

Our methodology is simple to state and strict to follow. People scope, direct and review every engagement. The platform executes only what has been approved, within the mode you chose. A finding is called real only when the evidence for its class is there. Our methodology is shaped by years of hands-on offensive research and responsible disclosure.

01 Lifecycle

The engagement lifecycle

  1. Scope

    Targets, exclusions, testing modes, reporting and deadlines, agreed with you in writing.

  2. Verify and authorize

    Written authorization, and domain ownership verified with a DNS TXT record or a verification file.

  3. Discover

    Assets and indirect signals, from public sources first, compared run to run.

  4. Test within the approved mode

    Scheduled and change-driven tests. Elevated steps wait for explicit approval.

  5. Prove and triage

    The evidence gate labels each finding; the team reviews every Signal.

  6. Report, advise, retest

    Signed reports, remediation guidance and a retest after each fix.

02 Roles

AI does the legwork. Experienced people direct it, correct it and answer for it.

Everyone is selling AI. We use it too, to cover more ground, more often. What you are really buying is judgment: a security director with a postgraduate degree in information security and cryptography, a master’s degree in cyber defense and more than 20 years of experience directs all of our security work, and the rest of the team monitors, follows and works every engagement.

What the platform does

  • Discovers your internet-facing assets and watches them for change
  • Reads signals outside your perimeter: certificate logs, leaked and infostealer credentials, public code, source maps
  • Runs approved tests on schedule and on change, only within the mode you authorize
  • Matches the technologies you run to new CVEs, with CISA KEV, EPSS and exploit context
  • Enforces scope and exclusions, and records every action in an append-only timeline

What our people do

  • Scope each engagement with you and verify domain ownership before any testing
  • Decide what gets tested, how deep, and when to ask for your approval
  • Triage every Signal, validate findings and correct what the machines get wrong
  • Prioritize by real-world impact and walk your engineers through the fix
  • Sign the report and advise your leadership
03 Inside the platform

Automated work is split into roles that check each other

Inside the platform, automated missions are divided into roles, each with a narrow job. The team directs them and reviews what they produce.

  • Planning

    Breaks objectives into waves: reduce uncertainty cheaply first, then examine high-value paths, then confirm or close each candidate finding.

  • Execution

    Performs only approved, policy-allowed work, checks scope and policy before each step and keeps raw outputs as evidence.

  • Evaluation

    Judges whether the evidence is sufficient or contradictory, and promotes an observation to a finding only when the evidence holds.

  • Prioritization

    Assesses how plausible an attack path is and how far the damage could reach. It never produces destructive playbooks.

04 Testing modes

Testing runs only in the mode you approve.

Four modes, from public sources only to operator-assisted testing that needs your explicit approval. No mode allows automated exploitation.

Testing runs only in the mode you approve.
ModeRiskWhat it can doWhat it never doesApproval
Passive analysisMinimalPublic sources only: DNS, WHOIS, certificate transparency logs, open-source intelligence, credential-exposure checksSends no requests to your systemsNot required
Safe validationLowAdds standard HTTP requests, TLS and header review, and TCP port checksNo vulnerability probes, no fuzzing, no authenticated testingNot required
Focused investigationModerateAdds endpoint probing, sensitive-path checks and technology fingerprintingNo fuzzing, no authenticated testing, no exploitationNot required
Operator-assistedElevatedAdds vulnerability probesNo exploitationYour explicit approval
05 Evidence model

The evidence model

Every finding carries a proof level. Validated means the evidence its class requires was captured. Signal means it was not, or not yet. Unproven findings are capped at medium severity, and critical severity requires exploitable or correlated proof.

The evidence model
Finding classEvidence required to be Validated
Server-side request forgeryA safe out-of-band callback or interaction
Open redirectA controlled Location header or final URL
Broken access control (IDOR and authorization bypass)An authorized versus unauthorized comparison
Authentication bypass and session issuesProof of a session-state change
CORS misconfigurationResponse headers obtained with a controlled Origin
Cross-site request forgeryA state-changing request and a token comparison
Unrestricted file uploadA safe upload and its retrieval
Path traversal and file inclusionA controlled, safe file read
Cross-site scriptingA payload reflection marker or proof of DOM execution
SQL injectionAn SQL error indicator, a timing difference over 500 ms, or a boolean-based difference
Default credentialsAn authenticated login with a safe test account
06 Prioritization

How we prioritize

Severity starts with CVSS and is weighed with CISA KEV (known exploitation), EPSS (likelihood of exploitation), public exploit availability and attack-path plausibility. Then comes the part no score captures: what the asset means to your business. That judgment is the team’s, and it is written down in the report.

07 Retest

Retest and change tracking

When you fix a finding, we test it again and record the result. Every run is compared with the previous one, so new, resolved and resurfaced findings are visible. A finding you closed that appears again is reopened, not silently ignored.

08 Boundaries

What we do not do

Never part of an engagement

  • Denial-of-service or load testing
  • Destructive actions, or modifying or extracting your data
  • Automated exploitation, in any mode
  • Testing outside the approved scope, or domains whose ownership is not verified

Never claimed

  • That a finding is real without the evidence its class requires
  • That a report is a certification or an attestation
  • That AI ran an engagement nobody was accountable for
  • Capabilities that are still on our roadmap
09 FAQ

Questions about the methodology

Who decides what gets tested?

You and our team, together, during scoping. The platform executes only what was approved.

Can automated missions exploit a vulnerability?

No. No testing mode allows automated exploitation.

How do you deal with false positives?

By requiring class-specific evidence before calling a finding Validated, and by having people review every Signal. No method removes false positives entirely; ours makes the uncertainty visible instead of hiding it.

Find it before an attacker does.

Tell us what you need covered. We will reply from contact@northboundintelligence.com to arrange a scoping call.

Facts on this page verified 11 November 2026.