An experiment in progress
Your systems,
checked by AI.
Sovereign GaiXer is an on-premises generative AI appliance that you install in your own office. The AI runs inside the appliance.
AI reads your source code one file at a time and looks for problems such as missing authentication checks and unvalidated input.
Systems keep changing, but reviews tend to happen only a few times a year.
ISSUE 01
Cloud AI and assessment services can check code quickly. But contracts or internal rules keep some companies from sending source code outside.
ISSUE 02
An expert assessment takes weeks from booking to report, and it is costly. You cannot order one every time the system changes.
ISSUE 03
Tools that look for fixed patterns flag many places that are fine. Real problems get buried while you sort through them.
An experiment that runs Clearwing, an open-source assessment tool (MIT License), with the AI inside Sovereign GaiXer to check source code for vulnerabilities.
It runs on Sovereign GaiXer.
Choose a repository, set the scope and run the scan. You can start with the parts that matter most, such as external entry points, authentication and payments. Progress and findings appear on screen as the scan runs.
The AI reads files one at a time and follows the flow of processing to find problems. Each finding includes the file, the location, the severity and a suggested fix, so developers can open the code and check it right away.
Examples of what it looks for
AuthSignatures or tokens are not verified; permission checks are missing
InputValues from outside are used in SQL, file paths or redirects without checks
FlowPayment or approval steps can be skipped; the same action can run twice
A fast AI goes through every file, and a careful AI checks the important files again. Each AI finds different kinds of problems, so layering them fills gaps that one AI alone would leave. Findings flagged by several AIs are marked for priority review.
Results can be exported as SARIF, Markdown or JSON. SARIF shows findings line by line in your code hosting tool, and Markdown can go straight into a report.
Your source code is read by the AI inside Sovereign GaiXer.
POINT 01
Source code is not sent to outside AI services.
POINT 02
Scans run in a zone that can reach neither the internet nor your internal network. Before each start, we confirm that nothing can get out.
POINT 03
Builds and reproductions run only in a disposable environment with no network, which is discarded afterwards.
DONE
Source code assessment now runs on Sovereign GaiXer.
Assessments use several AIs to check the code.
If this initiative interests you, please get in touch
Answers to common questions about this initiative
No. The AI runs inside Sovereign GaiXer, and assessments run in a zone that cannot reach the outside.
Please have someone check each finding against the code before acting on it. Some findings may already be prevented by earlier processing. Because each finding includes the file and location, checking takes little effort.
It depends on the method and the number of files. You can shorten it by checking the important parts first or by splitting the work across more appliances.
No. This experiment checks source code. It does not perform penetration tests on running servers or prove whether a problem can actually be exploited.