Cortex
An autonomous penetration testing console: AI agents carry out the steps of a penetration test, developed with AI models trained specifically for offensive security.
In development. Organisations interested in early access can contact us.
- taskrecon + port scan + record findings → acme-corp
- agentplanning the engagement, scope: 14 assets
- scanasset 09/14 · 1000 ports · tcp
- found203.0.113.7 · 22, 80, 443, 8080 open
- id8080 → Apache Tombcat 9.0.46 (out of date)
- thinkversion in the CVE window — queueing an exploit candidate
- findingHIGH · exposed manager console · evidence saved
- reportmapped to OWASP A05 · fix: update to 9.0.99
- done14/14 assets · 6 findings · 3m 06s
- cortex ›
From scope to proof, without a hand on the keyboard
Cortex runs the same method as our manual penetration tests — but the agents carry out each step themselves and decide where to go next.
- 01
Scope
You set what may be tested, in writing. Agents never step outside it.
- 02
Reconnaissance
They map the attack surface: assets, open ports, services, versions.
- 03
Exploit candidates
Findings that look exploitable are queued, ranked and checked — safely.
- 04
Evidence
Every finding is saved with the proof behind it, nothing is taken on trust.
- 05
Report
Each finding is ranked by real risk, mapped to OWASP and given a concrete fix.
- 06
Retest
After you fix, the agents run again and confirm what is closed.
The console
One place for every engagement: live risk, findings as they land, and the assets that need you first. Sample data.
What runs inside Cortex
Cortex is an autonomous penetration testing console by AngelForce, developed with AI models trained for offensive security. Coming soon.
- Autonomous agents
- Agents plan and carry out an engagement end to end, several assets at once, and keep going while you are away.
- Findings with evidence
- Each finding carries the request, the response and the reasoning behind it — ranked by real risk, mapped to OWASP.
- Exploit candidates
- Promising findings are collected, ordered by likelihood and checked within the agreed scope.
- Permission matrix
- A clear map of who can reach what, so broken access control shows up where auditors look.
- Tool and LLM logs
- Every tool the agents run and every model call is written down, so a run can be read back step by step.
- Reports and trends
- Findings become a report you can act on, with SLA and trend views, and a portal for the people who own the systems.
What Cortex is
- Autonomous
- AI agents plan and carry out the steps of a penetration test, from reconnaissance to the report, and choose each next step from what they have found.
- Trained for security
- Developed with AI models trained specifically for offensive security work, not with general-purpose chat models alone.
- From our practice
- Built by the team behind our penetration testing service, on the method we use in every engagement: scope, reconnaissance, exploitation, reporting and retest.
- Evidence and fixes
- The aim is the same as in our manual tests: findings with evidence, ranked by real risk, each with a concrete fix.
Authorised use only
Cortex is made for testing systems you own or are allowed to test, within a scope agreed in writing. Users accept its terms of use when they sign in.
We will share more about Cortex when it is ready. Until then, our penetration testing team tests your systems by hand.
Put Cortex on your own perimeter
We are opening Cortex to a small number of organisations before launch. Tell us what you want tested and we will see whether your systems are a fit for the pilot.
Ask about early access