How it works
Two tracks, one standard of evidence.
The AI your staff use and the AI you build carry different risks, so each track has its own method. Both share the same rules: scope agreed first, testing only under your signature, and every finding shown with evidence.
The two tracks
Different risks, different methods.
Track one
The AI your staff use
The risk: company data leaving through tools you do not control, kept by vendors, or used to train their models.
- Find it: the Shadow AI Review ($2,500)
- Control it: AI Policy & Controls ($4,500)
- Watch it: AI Security Monitoring (monthly)
Track two
The AI you build
The risk: your chatbot, assistant, or agent manipulated into revealing data, breaking its rules, or taking actions it should not.
- Test it: the AI Red-Team Assessment (from $7,500)
- Retest it: one retest included
- Watch it: quarterly retests with monitoring
Who does what
You authorize. We test and report. Your team fixes.
- 01
Your system owner
The executive who authorizes the review or test.
- Signs the scope and authorization
- Decides which findings to fix, and when
- Accepts any remaining risk
Never surprised by what is tested.
- 02
Custos engineers
Sophrono engineers who build AI systems.
- Review tools, data, and vendor terms
- Test AI systems within the agreed rules
- Report every finding with evidence
Never test outside the signed scope.
- 03
Your team
IT administrators and the engineers who own the systems.
- Provide access and a test environment
- Fix findings and configure controls
- Join the readout
Never left with a finding they cannot reproduce.
Rules of engagement
Signed before any test begins.
Every red-team assessment starts with a document you sign.
- ScopeThe systems, components, and environment in scope, and what is out of scope.
- MethodsThe test areas and techniques, and any that are excluded.
- LimitsDates, hours, rate limits, and data that must not be touched.
- ContactsWho to call for a critical finding, and who can pause the test.
The report
What it says, and what it does not.
- What was reviewed or tested, by which method, and when
- Every finding with evidence, a rating, and a specific fix
- What was out of scope, so nobody mistakes it for covered
- For red-team work, the status of each finding after the retest
After the call
What clients usually do next.
The services on each track, from lightest to most hands-on.
- 01Check, freeFreeTake the five-minute shadow AI check, or ask a security engineer about your tools or system.
- 02Find the exposure$2,500Every AI tool your staff use, the data reaching it, vendor terms, and prioritized fixes.
- 03Control it, or test what you build$4,500 / from $7,500A policy with enforced settings for staff AI use, or a red-team assessment of the AI you build.
- 04Keep watchMonthlyMonthly checks of new tools and settings, and quarterly retests of the AI you build.
Each step stands on its own. You never commit to the next one in advance.
Questions
Why two tracks?
The risks are different. Staff AI use is about data leaving the company through tools you do not control. AI you build is about your own system being manipulated. Each needs a different method.
Do you test production?
Staging by default. Production testing only with explicit written approval and agreed limits.
What happens if you find something serious during a test?
The rules of engagement name a contact. Critical findings are reported to them the same day, not held for the report.
Can we fix findings ourselves?
Yes. Every finding comes with a fix your team can make. We can help, but you are free to use anyone.
Choose a track