Autonomous pentesting, with guardrailsOffensive by design. Careful by default.
Escape AI re-tests your findings on your live app the way a pentester would. Here are the checks, limits and records around each re-test.
The guardrailsWhat keeps each scan in check.
- 01
Scope
Targets outside your allowed hosts, and private or internal addresses, are refused when you save them and when a scan starts. During the scan, allowed hosts are rules the agents are told to follow.
- 02
Fresh container per scan
Each scan’s tools run in a fresh, resource-limited container. It’s shut down when the scan ends and can’t reach private or internal networks.
- 03
Health check
For re-tests started by a merge, Specurity waits the delay you set and, if you add a health URL, until your app answers.
- 04
Verdict and evidence
Every finished re-test ends with a verdict: Reproduced, Not reproduced or Inconclusive. When the issue reproduces, you also get the evidence, a proof-of-concept description, the impact and steps to fix.
- 05
Audit log
Logins (including failed ones), changes and downloads are logged. Each Run keeps a log of what the agents did.
- 06
Watch and stop
Watch a re-test’s activity live and stop it at any time, whether you started it with Run on the issue or a merge did.
Scope and accountsYou set where it goes and what it can use.
- Allowed hosts, checked twiceYou list targets per environment. Hosts outside your allowed list, and private or internal addresses, are refused when you save targets and when a scan starts.
- Rules the agents followDuring the scan, allowed hosts, excluded paths, the rate limit and read-only are rules the agents are told to follow.
- Test accounts by roleAdd several test accounts by role, including two-factor. Escape AI re-tests business-logic and access-control findings too, signing in with your test accounts per role.
- Encrypted credentialsCredentials live in Creds Vault, encrypted and hidden in lists. Only people the vault allows can reveal a value, and every reveal or copy is logged.
- GitHub access you chooseRead-only access to code and pull requests in the repos you choose, plus a webhook so we hear about merges.
- Skills, reviewed by a teammateSkills teach the AI how your app works. A different team member reviews every skill change before it’s used.
In detailRe-tests after a merge, step by step.
It can wait until your app answers.
When a project re-tests on merge, Specurity waits the delay you set after the merge. If you add a health or readiness URL, it also waits until your app answers, then re-tests the issue.
- Off until you turn on re-test on merge
- Delay you set: 5 minutes by default, up to 24 hours
- Optional health or readiness URL
- If the fix didn’t hold, the issue goes back to the team

Data and recordsYour data stays apart. Logins, changes and downloads stay on record.
Data kept apart
Project data is separated per company and visible only to members you give access to. Our databases, files and scan servers are in AWS Mumbai; some sub-processors, such as our AI model provider, process data outside India.
An audit log
Logins (including failed ones), changes and downloads are logged. Each Run keeps a log of what the agents did.
Encrypted credentials
Values in Creds Vault are encrypted and hidden in lists, revealed only to people the vault allows. Every reveal or copy is logged.
Access you control
8 roles from Owner to Guest, with permissions per project and per credential vault, and an option to require 2FA for everyone.
- SOC 2 in progress
- Data stored in AWS Mumbai
- 2FA enforcement
FAQSafety questions.
Can the agents test anything outside my scope?
Targets outside your allowed hosts, and private or internal addresses, are refused when you save targets and when a scan starts, and scan containers can’t reach private or internal networks. During the scan, allowed hosts, excluded paths, the rate limit and read-only are rules the agents are told to follow.
Can other customers see my data?
No. Project data is separated per company and visible only to members you give access to. Each scan runs in a fresh container that is shut down when the scan ends.
Can I stop a re-test?
Yes. Watch its activity live and stop it at any time, whether you started it with Run on the issue or a merge did.
Does a re-test wait for my deploy?
A re-test started by a merge waits the delay you set (5 minutes by default, up to 24 hours) and, if you add a health URL, until your app answers. Re-test on merge is off until you turn it on, and you can always start a re-test with Run on the issue.
Which accounts does the AI use?
The test accounts you give it, by role and including two-factor. Escape AI re-tests business-logic and access-control findings too, signing in with your test accounts per role.
What does a re-test give me?
Every finished re-test ends with a verdict: Reproduced, Not reproduced or Inconclusive. When the issue reproduces, you also get the evidence, a proof-of-concept description, the impact and steps to fix.
Where are my test credentials kept?
In Creds Vault. Values are encrypted, shared per vault and hidden in lists. Only people the vault allows can reveal one, and every reveal or copy is logged.
Where is my data stored?
Our databases, files and scan servers are in AWS Mumbai (ap-south-1). Some sub-processors, such as our AI model provider, process data outside India. SOC 2 is in progress; the report is not yet issued.
RelatedKeep exploring.
Re-test your fixesSee how a fix gets re-tested.
Bring a recent pentest report. Specurity reads the findings, so nobody retypes them.