Security
AI does the work.
It does not own the system.
TAZM is built on a small set of security principles that decide how AI work is run, how credentials are handled and when a person must decide.
Principles
How we approach security.
Controlled execution
AI work runs within limits set by TAZM, not limits the AI sets for itself.
Least privilege
Work is given the access the task needs, and no more.
Separated credentials
Credentials are kept apart and used only where a task requires them.
Fail closed
When something goes wrong, the safe outcome is to stop, not to carry on unchecked.
Your approval for sensitive actions
Changes that matter — such as security, credentials, spend or deployment — wait for your decision.
Auditable evidence
Important actions and results leave a record you can review.
Tested, not assumed
Security controls are tested by trying to break them.
Adversarial testing
We test our security controls by actively trying to defeat them, and we keep the results.
Re-checked on change
When the architecture changes, the affected controls are tested again.
More detail
Security documentation for customers.
On request
Detailed security information is available to customers and prospective customers on request, under a confidentiality agreement.
Build with AI.
Keep control.
Security principles that protect the work without getting in its way.