Leakage
Sensitive fields reaching external models in identifiable form. SyniSense refuses the call before it leaves the perimeter.
The layer that lets sensitive data work with frontier external models without leaving the perimeter in identifiable form. Anonymises on the way out. Re-identifies inside the perimeter on the way back. Produces an audit receipt for every call.
Our compliance officer signed off. The model never saw a real name.
Every field replaced, every token used, every mapping reference — recorded against the request.
The exact payload the external model received, in pseudonymised form. Re-derivable for audit, never re-derived in production.
The model response before and after re-identification, with the diff. Inspectable by compliance, security, and data protection teams.
Sensitive fields reaching external models in identifiable form. SyniSense refuses the call before it leaves the perimeter.
Model output that infers identity from quasi-identifiers — postcodes, dates, rare combinations. SyniSense flags reconstructable patterns.
Requests that cannot be replayed under audit. SyniSense refuses calls that cannot produce a complete receipt.
Boundary policy that has slipped out of alignment with regulation or internal control. SyniSense re-validates policy on every call, not on every deploy.
When the Akki platform is deployed, SyniSense is the data boundary layer. Every external model call is mediated through it. Audit receipts are written to the platform's own observability stack.
For organisations using Copilot, Claude, or GPT through existing infrastructure, SyniSense deploys as a perimeter SDK. The mapping store stays inside the customer's environment; nothing about the perimeter leaves the perimeter.
Deployment begins with a conversation about the perimeter, the data classes inside it, and the regulators who will read the receipts.