Home / platform
Our platform

RedCore: our own AI, running on GPUs we own in Canada.

Four specialised AI assistants, our own private GPU servers in Canadian data centres, and a human expert in charge of every important decision. This is the technology behind every Red Intelligence security test.

01 / How the AI works

Four AI assistants, one job

Each assistant is trained for one task. They pass work to one another, and our human expert can see everything and step in at any point.

AGENT 01

SCOUT

Finds everything that could be attacked

Quickly lists your websites, systems, accounts and software, and builds a map of everything an attacker could target.

result → map of your systems
AGENT 02

ARCHITECT

Picks the most likely ways in

Sorts thousands of possibilities by business impact and likelihood, ignores false alarms, and suggests the realistic multi-step attacks worth testing.

result → ranked test plan
AGENT 03

PHANTOM

Tests safely and records proof

Runs only the steps a human has approved, within strict safety limits on speed, targets and impact. Every action is logged so it can be replayed.

result → confirmed proof
AGENT 04

SCRIBE

Writes the first draft of your report

Turns confirmed issues into a clear business-risk summary and step-by-step fixes for your IT team. Then a human editor makes sure every word is true.

result → draft report

SCOUT → ARCHITECT → PHANTOM → SCRIBE → HUMAN SIGN-OFF

02 / Private by design

Your data stays under our roof

INFRA-01

Our own GPU servers, in Canadian data centres

Our AI runs on GPUs we own and physically control, in Canadian data centres. We don't use big-tech AI services or share machines with strangers, and your data stays in Canada — written into your contract.

INFRA-02

Every client kept separate

Each test gets its own sealed-off, encrypted space. One client's information can never influence or leak into another's results.

INFRA-03

Deleted when we're done

Everything from your test is securely destroyed 30 days after we deliver the report. You can witness the deletion, and every step is recorded in a tamper-proof log.

INFRA-04

Locked-down internet access

Our test environments can only reach the systems you've approved in writing. Outside AI services are blocked at the firewall, by design rather than by policy.

INFRA-05

Fully offline option

Government, critical infrastructure and highly regulated clients can run everything on a disconnected system at their own site. Same process, same quality, no outside connection.

INFRA-06

Audited annually

Independent auditors check our systems and processes every year (SOC 2 Type II and ISO 27001). Reports are available under NDA during procurement.

03 / Humans in charge

The AI suggests. Our experts decide.

The AI can suggest a way in, but a certified expert approves each step, watches it run against your live systems, and records the proof. The AI drafts the report, but a person checks every claim. That's why fewer than 2% of our findings are false alarms, and why nothing in your report is a guess.

  • ✓
    Nothing risky without approvalAny step that could change or affect your systems needs an expert's sign-off first.
  • ✓
    A full record of what the AI didAsk for it any time: what was tried, what was rejected, and why.
  • ✓
    Certified experts on every testOSCP, OSWE, OSEP and CREST-certified people review everything before it reaches you.
04 / Your report

Reports your leadership and your IT team both trust

An executive summary up front, full technical detail underneath, and fixes your developers can start on the same day. Here's a sample page with client details removed:

RI-2026-0774 · Web & API Penetration Test · Client: [redacted]PAGE 12/48

RI-0774-003 · Chained IDOR leading to tenant-wide privilege escalation

CVSS 9.1 CRITICAL · OWASP: A01:2021 · verified by operator M.K. · 2026-10-14 09:32 UTC
CRITICAL

Business impact: any logged-in customer could download every other customer's invoices, including the company's own administrator billing data.

The /api/v2/invoices/<id> endpoint returns invoice objects for any numeric identifier without verifying tenant membership. Combining this with the bulk-export feature's missing role check allows any authenticated user to enumerate and export all tenant invoices, including those of administrative users.

Impact: cross-tenant data exposure and full read access to billing data. Chained with finding RI-0774-001, write access is achievable (see §4.3).

GET /api/v2/invoices/8f3a-…?expand=all HTTP/1.1 Authorization: Bearer [standard-user-token] HTTP/1.1 200 OK { "tenant": "acme-corp-admin", "total": "184,203.00", … }

Remediation: enforce object-level authorization in the invoice repository layer (defense in depth — not in the route handler), and add a regression test asserting cross-tenant 403s. Retest included within 90 days.

05 / Works with your tools

Issues go straight to your team's tools

Jira ServiceNow GitHub GitLab CI Slack Microsoft Teams Burp Suite SARIF

Every issue is also delivered in formats your IT team's software can read (SARIF, JSON, CSV), so fixes are tracked in the tools they already use, not in a PDF.

Want the full technical details?

We share our architecture overview, data-flow diagrams and audit summaries under NDA during procurement. Ask for the "Platform Trust Pack".