Industry offer · Uniformed services

AI for the police and uniformed services

SAIE for uniformed services is an AI environment with local models in which service information does not go to external services, and the commander sees who had access to what. AI under the commander’s control: local models, voice reports, a full access trail

Why now

Service information must not be processed in external services, notes and reports take time in the field, and procedures are extensive while time to check them is short. You must show who had access to which information, and the autonomous operation of agents must remain under the commander’s control. Since April 2026, an AI team has been operating at MSWiA (Ministry of the Interior and Administration), and from 28 October 2026 KRiBSI (Commission for the Development and Security of Artificial Intelligence) can inspect the use of AI. AI in law enforcement and border control is a high-risk system — AI Act obligations will cover it from 2 December 2027

Three people who need to say “yes”

Chief / commander — control over what AI does on their behalf

Head of communications and IT — isolation from the public network, local models

Information protection officer — access trail and agent register

Business scenarios

Notes and reports by voice

Process automation

TWIN:DESK

Problem: Writing notes and reports takes time in the field

How it works: Voice chat (STT/TTS) and structured answers that put the content into the set report template

What you gain: The report is dictated and structured in the template while still in the field — to approve, not to rewrite

Assistant for procedures and service instructions

Knowledge and documents

TWIN:DESK

Problem: Procedures and service instructions are extensive, and time to check them is short

How it works: A central RAG knowledge base that cites the relevant regulation or procedure

What you gain: The officer gets the right regulation with a citation in seconds — without searching through extensive instructions

Quick onboarding of users to work with AI

Adoption and skills

CDF

Problem: Officers have no time for long training on new tools

How it works: CDF Academy (F3) — basic and specialist tracks per position, with internal certification

What you gain: Short training tailored to the position ends with a certificate — the officer works with AI safely from the first week

Sovereign AI for service information

Security and sovereignty

PROXY:AI + CDF

Problem: Service information must not be processed in external services

How it works: Sovereignty Level L4 (CDF-F1) — isolation from the public network; local models only, offline mode

What you gain: AI runs in the service network with no connection to external services — in line with information protection rules, not against them

Access control and a full audit trail

Compliance and oversight

PROXY:AI

Problem: You must show who had access to which information and when

How it works: ABAC access decisions based on role, unit and data type; a cryptographically signed query log

What you gain: Every access to information leaves an immutable, signed trace — showing “who, what, when” takes a moment

Controlled operation of AI agents

Compliance and oversight

TWIN:DESK + CDF

Problem: The autonomous operation of AI agents must remain under the commander’s control

How it works: An agent register with autonomy levels and an emergency shutdown procedure (CDF-F2); rollback of agent actions with a visible rollback plan and confirmation

What you gain: The commander knows which agent is running, at what autonomy level, and can stop it or roll back its actions

Evidence

A variant cut off from the network (full network isolation) and offline operation

A cryptographically signed log with retention of up to 2 years

How we start

A demonstration in an isolated environment on unclassified service instructions — a voice report and an access trail

Questions decision-makers ask

How do you get the use of unauthorized AI tools under control?

Not with a ban, but with a safe tool, a policy and training. PROXY:AI routes traffic to approved models, masks data and records queries, and training teaches staff what must not be entered

Shadow AI

How can you limit hallucinations?

By answering only from service documents, with a source citation, and by checking the output before use

For which document tasks is AI safe?

For summaries, searching procedures, and drafting notes and reports — always with human verification. Not for autonomous decisions concerning individuals

All questions