Autonomous testing that can't leave its lane.
Penetration testing is slow and the people who do it well are scarce. Autonomous agents can close that gap, but only if safety is part of the architecture rather than a setting. That's what we're building.
Every action passes the gate. Every decision is on the record.
An autonomous agent that can attack is only useful if you can prove it attacked nothing else. The scope gate sits on the single path every action must take, and it fails closed.
Eight controls in the core.
Six are proven in our private lab against practice applications. None has run against a real customer system yet.
Scope check
Runs before every tool launch. Targets outside the agreed scope are refused before any traffic is sent. It can't be switched off.
Rate control
Every host is paced, so testing never floods your systems.
Independent re-test
Each finding is confirmed by a separate automated re-test against the target before it reaches a report.
Signed evidence
Evidence is cryptographically signed, so any tampering shows.
Decision log
Every decision goes into a log that can't be edited or deleted.
Reversible cleanup
Changes made during testing are undone; anything that can't be is quarantined and reported.
Autonomy modes
Unattended runs, which need your recorded consent, are proven in the lab. Supervised and dry-run modes are built.
Tenant isolation
Role-based access, and each customer sees only its own data. Built and tested, not yet lab-proven.
One coordinator. Specialist agents.
A lead AI agent plans the test and hands work to specialists, who can only read until a change is explicitly allowed. The coordinator and the web, API, infrastructure and network specialists are built; they aren't switched on yet. Mobile, cloud, client-side and AI specialists are in progress.
Honest status.
We'd rather you trust our roadmap than be impressed by it. No dates until we can keep them.
Running in our lab
- Unattended tests of practice web applications
- Scope and rate checks on every tool launch
- Reconnaissance and vulnerability scanning end to end
- Findings confirmed by an independent re-test
- A signed evidence bundle and a log that can't be edited
- Tools run isolated, with no internet access
Built, not yet lab-proven
- Supervised and dry-run modes
- Role-based access and tenant isolation
- Read-only specialist agents for web, API, infrastructure and network
- A signed, machine-readable report
In progress
- An AI coordinator that runs tests unattended (built, not yet switched on)
- Specialists for mobile, cloud, client-side and AI systems
- Sandbox rules for every tool launch
- An emergency stop
Next
- More live tools beyond our current scanners
- Vulnerability discovery at scale
- Per-target planning and exploit validation
- Readable customer reports and a hosted multi-tenant service
Lab means our isolated test network with deliberately vulnerable practice applications.
Want to know when it's ready?
Register your interest and we'll get in touch when there's a build worth putting on a real scope.
- Who it suits
- Teams that test oftenIn-house security, consultancies, AI-native products.
- What you'll hear
- Honest progressWhat each build can and can't do, when it's ready.
About the platform.
Can I use it today?
Not yet. It's a working prototype: in our private lab, against deliberately vulnerable practice applications, it has run a complete unattended test with every safety check in place. It hasn't been run against a real customer system, and it isn't hosted yet. We'll tell you plainly what each build can and can't do.
Will it replace human testers?
No. It's built to cover the repetitive breadth of a test so people can spend their time on judgment: chaining findings, proving impact, and deciding what matters to your business.
How is it different from a vulnerability scanner?
A scanner matches known signatures and stops. The platform is built to work through a whole engagement the way an operator does, checking scope before every action, pacing its traffic and re-testing each finding, with evidence that can't be quietly altered.
Does AI decide what gets reported?
Not today. Findings come from real tool output and are confirmed by a separate automated re-test, not by an AI model. The AI coordinator is built but not yet switched on.
How do I get early access?
Register your interest. There's no programme to join yet; we'll contact registered teams when there's a build worth putting on a real scope.
What will the product be called?
We'll announce the name when its trademark is filed.
Hear first when it's ready for a real scope.
Security teams, technology partners and investors interested in governed autonomy: register interest and we'll be in touch.