Your data stays your data.
Enterprise AI on commercial terms. Your information is never used to train models, it stays isolated to your tenancy, and every claim on this page is one we will put in writing.
What is true today
In productionThese are controls that are in place now, not roadmap items. Where something is still on the path, we say so on this page rather than leaving you to find out later.
Your data never trains a model
Client data is analyzed, reported, and left in place. It is never used to train shared or public models, ours or anyone else's. This is a contractual commitment, not a setting.
Isolated by tenant
Each customer's data and configuration run in their own tenancy. Nothing crosses into a shared engine or another customer's environment.
Server-side only access
Credentials and privileged operations stay server-side. The browser never holds the keys to your data.
Tamper-evident records
Each analysis is committed to a hash chain. A later change to a record breaks the chain, so tampering is evident rather than silent.
Deterministic where it matters
Rules that should be deterministic are deterministic, each with a stable identifier. The same inputs produce the same output, which is what makes a result reproducible in review.
Disclosed inference
Where a platform infers rather than reads, it records that it inferred, and why. Uncertainty appears in the output instead of hiding inside it.
On the path
UnderwaySOC 2 Type II
Underway. We will publish the report when it exists rather than describing it before it does.
FedRAMP
On the path, driven by our federal and research-security work.
A named security lead
Our security program is led by Dr. Marian Zaki, Director and lead security architect. She owns the threat model and the security architecture behind every platform, and she is the person your security team will talk to.
What we will send you
Architecture overview, data-flow description, tenancy and access model, subprocessor list, and our answers to your security questionnaire. If something is not yet in place, the answer says so.
The technology is our own.
Every product runs on the same defensibility layer, built here — how an answer is tied to its source, how the work is made checkable, how your data is kept separate from everyone else's. That layer is covered by fourteen U.S. patent applications, which is why every platform in the portfolio carries Pat. pend.
Send us your questionnaire.
We would rather answer your security team's questions early than late.