AI in the military and defence industry — air-gapped

Assistant for documentation, standards and technical requirements

Knowledge and documents

Problem: Documentation, standards and technical requirements are extensive, and time for analysis is short

How it works: A central RAG knowledge base that runs in an environment cut off from the network, with OCR and source citations

What you gain: Analysis of technical requirements starts with an answer and a citation, not with reading a thousand pages

AI in an environment completely cut off from the internet

Security and sovereignty

Problem: Work with classified information rules out any internet connection

How it works: Sovereignty Level L4 (CDF-F1) — full network isolation (air gap); local models only; offline mode with local integrations

What you gain: The full functionality of an AI environment without a single packet leaving — in line with the classified information protection regime

Sovereignty and risk assessment of AI suppliers

Security and sovereignty

Problem: The geopolitical risk of AI component suppliers must be assessed and documented

How it works: A sovereignty matrix (CDF-F1) assessing suppliers from the US, China and Europe; Sovereignty Level Assessment; Model SBOM and Cyber Resilience Act compliance reporting; model validation before launch

What you gain: Every AI component has a documented origin and risk assessment — the supplier decision can be defended before auditors and superiors

Controlled operation of AI agents

Compliance and oversight

Problem: The autonomous operation of AI agents must remain under operator control

How it works: An agent register with an explicit autonomy level, permission limits and an emergency shutdown procedure (CDF-F2); DAG-based rollback of agent actions

What you gain: The operator knows what each agent does and can shut it down immediately or roll back its actions according to a visible plan

Access control and a full audit trail

Compliance and oversight

Problem: You must show who had access to which information

How it works: ABAC access decisions; a cryptographically signed, immutable log with retention

What you gain: “Who, what, when” is in a signed log — showing it takes minutes, not days

Competence check before approving a decision

Compliance and oversight

Problem: Operators approve system recommendations without fully understanding the consequences

How it works: Human Competence Gate (CDF-F2) — check questions before a critical decision is approved; CDF Academy — supervisory track

What you gain: Approving a system recommendation requires showing understanding — human oversight is a fact, not a click

Are AI decisions traceable?

Yes. Every query and answer goes to the WORM log in PROXY:AI, and TWIN:DESK answers include a source citation

Does the solution ensure control over the technology?

Yes. Models run locally, code and configuration stay with the client, and the list of approved models and components is under the operator’s control

Does it work in an isolated server farm?

Yes. PROXY:AI with local models runs in air-gap mode, without an internet connection. We carry out deployments in classified environments together with integrators authorized for such work