Trust, security, and AI you can audit
Switching the systems that run your operation is high-stakes. This page lays out how FlightAtom protects your data and governs its AI: what is live today, and what we will share under NDA during procurement.
What protects your data today
Encryption in transit
Operational data is encrypted in transit. Workflow-appropriate access controls are scoped to the modules and record types included in the implementation.
Least-privilege access
Where implemented, role-based access can be scoped by module, record type, and responsibility according to the agreed authorization model.
Your data stays yours
Customers retain ownership of their records and can export supported data according to the agreed implementation and data-handling policy.
On-premise Delphi option
Delphi can be configured for an on-premise deployment path when the agreed architecture and model setup must keep document and query data within the operator's network.
Reviewable records
Aurora and its connected add-ons preserve workflow-appropriate records and evidence for review across safety, training, crew, and other configured areas.
Modular scope and exports
Start with Aurora, add focused products when a workflow needs more depth, and define supported data exports and connection boundaries in the implementation.
The procurement checklist
Enterprise buyers need specifics before procurement. The rows marked On request are answered in detail in our security documentation, shared under NDA during evaluation. This public page is updated as each is finalized.
Four kinds of “AI”: named, not blurred
Calling everything “AI-powered” makes technical buyers more skeptical, not less. FlightAtom keeps these four layers distinct, and a human owns the decision that matters.
Compliance & legality rules
Configured duty, rest, FTL, and qualification logic can be represented as deterministic rules rather than predictions. Operators validate rule scope, source data, and exceptions.
Predictions & scoring
Fatigue scoring, anomaly detection, and risk trends are statistical models. They identify where to look and provide a score, not a verdict.
Cited, grounded answers
Delphi retrieves passages from configured document sources and composes answers with citations. Source scope and access depend on the implementation, and users must verify the cited evidence.
Decisions & approval
People approve rosters, sign the dispatch release, and own the call. Where configured, software can propose and explain; a qualified human decides.
How Delphi stays grounded
Delphi retrieves from an operator-provided document corpus and can show section- and page-level source context, confidence or groundedness states, and an abstention when configured evidence is insufficient.
- Searches the authenticated document corpus configured for the operator; open-web search is not assumed.
- Supported answer views can show source details such as document title, section, and page.
- The configured workflow can expose confidence or groundedness context, ask for clarification, or abstain when evidence is weak.
Benchmark figures for answer accuracy, citation precision, and abstention rate, measured against a defined document set and methodology, are shared under NDA during evaluation.
Need the full security pack?
We share detailed security documentation, our DPA, and architecture diagrams under NDA during evaluation.
Request security documentation