Trust center

An enforcement companyshould expect to be examined.

The external evaluation is complete. Its approved scope, reviewed version and findings will be published here when the public summary is available.

External evaluation complete

Independent review

Adversarial review,paid for by us.

Phase 1 · Architecture
External cryptography practice
Complete
The approved public summary, reviewed version, date, scope and reviewer attribution are awaiting supplied material. Completion is not a claim of certification.
Specifications
Reviewable under NDA
The full specification bundle, dozens of formal specs under version control with a zero-error lint baseline, is available to qualified reviewers and customers under NDA.
Reproducibility
Two implementations, one hash
Independent kernel implementations must produce hash-identical decision records on identical inputs. Any divergence between them is handled as a defect.

Architecture posture

Fail-closed
Denied by default
Unrecognized proposals, degraded dependencies, and configuration drift all resolve to denial, because the kernel has no permissive fallback to fall into.
Default-deny
Only allowed actions occur
The kernel admits a proposal only by its membership in a pre-vetted schema, which means free-form instructions have no execution path to find.
No model authority
Zero standing credentials
Models hold no keys, tokens, or write paths at any point, and authority is exercised only inside the attested boundary, only by the kernel.
Attested custody
Hardware-signed decisions
Every decision is signed inside a hardware-attested custody chain, and verifying that chain requires no trust in Instantial or in any model vendor.

Data handling

Records
No personal data in records
The record schema is incapable of carrying personal data, which means there is no scrubbing step that anyone can get wrong.
Retention
Append-only, tamper-evident
Records are append-only and hash-chained, so any alteration is detectable by whoever holds a copy of the chain.
Training
No training on your data
Instantial does not train models, and customer data is never used to improve anyone else's.

Provenance

Intellectual property
Patents filed
Core enforcement and attestation methods are the subject of filed patent applications, prepared and prosecuted by outside counsel.
Responsible disclosure
Report a vulnerability
Security reports go to security@instantial.ai. We acknowledge reports within two business days, and we do not pursue researchers acting in good faith.
Compliance
Certifications roadmap
On the roadmap
Formal certifications (SOC 2 and successors) are on the roadmap and will be listed here as they are achieved. We will not claim them before then.