Scenario 1
Your company is planning to adopt AI
We assess processes and infrastructure, identify suitable use cases, design the architecture and build a secure AI environment — from a pilot on real tasks to production operation.
Design enterprise AILEGEND-A · an ITBP JSC solution
AI servers and enterprise solutions
We supply AI servers and design and deploy enterprise AI systems. Workstations, GPU servers and infrastructure for local models, RAG, AI agents and machine learning workloads.
Your organisation retains control of data, access rights and AI actions.
Local deploymentAccess controlAI action auditingData leakage controlIntegration with enterprise systems
01 · Infrastructure
AI is becoming part of enterprise infrastructure
Companies are adopting enterprise LLMs, RAG, AI agents and automation tools. At the same time, new access points emerge for documents, correspondence, databases, business systems, APIs and enterprise accounts.
Connecting models and agents to information systems creates a new interface to enterprise information. Enterprise AI should be treated as a complete information system, with an owner, boundaries, permissions, logs and security requirements.
Target architecture for an enterprise AI environment
Simplified diagram
Employeesusers, managers, security and IT teams
SSO / IAMauthentication, roles and groups
Secure environment
LEGEND-A / AI Gatewaysingle entry point, model routing and external LLM control
Policy / Security LayerRBAC, data policies, filters, DLP and prompt injection protection
LLM · RAG · AI Agentslocal models, knowledge bases and agent tools in an isolated environment
1CERPCRMDocument managementEmailFilesDatabasesAPIRPA
Single entry pointSecure environment boundaryComponents in the customer's infrastructure
Layer 1
Functional AI environment
- Local and enterprise LLMs
- Enterprise knowledge bases and RAG
- AI agents and multi-agent scenarios
- Orchestration and workflows
- Integrations: 1C, ERP, CRM, document management, email, databases and APIs
- RPA where APIs are unavailable; MCP for tools
Layer 2
Information security environment
- AI Gateway and external LLM control
- Access management, RBAC and SSO
- Logging and auditing of AI actions
- DLP, leakage control and SIEM
- Agent isolation and secret protection
- Anomaly monitoring and investigations
This approach embeds AI in real business processes while retaining control over enterprise data, user permissions and agent actions. Enterprise AI security is designed alongside its functionality, rather than added after deployment.
02 · Solutions
Two ways we work with customers
ITBP JSC designs and deploys enterprise AI, assesses existing implementations and their architecture and security, establishes a secure AI environment, and provides control over data, permissions, agent actions and external connections.
Scenario 2
AI is already in use
We inventory AI services and assess data flows, architecture, access and the security of existing AI systems. We then propose a target secure architecture and a risk mitigation plan.
Request an audit03 · Enterprise knowledge
Shared information context for employees and AI agents
Enterprise documents, correspondence, regulations, contracts, decision history and information from business systems form a managed knowledge base. The model answers using the organisation's sources and cites them.
Users and AI agents receive only the information they are authorised to access. Permissions are inherited from source systems and checked on every request.
Source 01Contracts and appendices
Source 02Regulatory documents
Source 03Technical documentation
Source 04Internal policies
Source 05Primary accounting documents
Source 06Payments
Source 07Decisions and their rationale
Source 08Counterparty history
Source 09Projects
Source 10Requests and enquiries
Source 11Digitised paper archives
Source 12HR documents
Source 13Internal system data
Source 14Enterprise email
Source 15File repositories
Source 16Databases and APIs
Indexing
Managed knowledge base
Connect sources, chunk and vectorise documents, and update indexes on a schedule or in response to events. Document versions and currency are tracked.
Access
Document-level permissions
Each answer uses only documents available to the specific user or agent. Source-system restrictions remain in force in the knowledge base.
Verifiability
Answers with source references
Employees can see which documents and passages support an answer and open the original in its source system.
04 · AI agents in business processes
From information to action
AI can both find information and participate in business processes through authorised tools: business systems, document management, email, APIs, RPA workflows and managed tool interfaces such as MCP.
Each agent operates within its assigned role, with the minimum necessary permissions and a complete action log.
Finance
Finance function
- document analysis
- reconciliation with accounting systems
- report preparation
- discrepancy detection
Legal
Legal function
- contract analysis
- identifying obligations
- deadline monitoring
- drafting documents
Procurement
Procurement operations
- proposal analysis
- comparing terms
- documentation analysis
- preparing supplier summaries
IT
IT and support
- request handling
- knowledge base search
- incident classification
- routine task automation
Management
Leadership
- analytical briefings
- management summaries
- finding connections
- working with enterprise context
Information security
Information security
- event and log analysis
- working with regulatory documents
- AI system oversight
- incident investigation
Critical actions can proceed only after confirmation by an authorised employee. Posting accounting documents, making payments, changing permissions, sending data outside the environment and other irreversible operations require explicit human confirmation.
05 · Secure environment
AI within your company's infrastructure
Enterprise documents and knowledge bases can be processed by local models without sending their content to an external AI provider. Deployment depends on the data, security requirements and the customer's computing resources.
Option A
Local environment
LLMs, RAG and enterprise data run in the customer's infrastructure: a data centre, company servers or dedicated GPU nodes. Data stays within the organisation's perimeter.
Option B
Isolated environment
Operation without public AI services and, where required, without Internet access. Models and components are updated through controlled customer procedures.
Option C
Hybrid environment
Sensitive data is processed locally, while authorised requests to external models pass through a controlled AI Gateway with policies, filtering and logging.
Models
Local and enterprise LLMs
We select models for the customer's tasks and hardware, including Russian and open models. A single access interface works regardless of where the model runs.
Infrastructure
Computing resources
GPU capacity, storage and network sizing for the workload. Deployment in existing virtualisation environments or on dedicated hardware.
Operations
Manageability
Monitoring, redundancy, model and index updates, operating procedures and knowledge transfer to the customer's team.
06 · Enterprise AI security
Enterprise AI must remain manageable
Models, knowledge bases and agents access the same data as employees. They therefore require the same safeguards: access control, information-flow control, logging and monitoring.
01 · Access control
AI does not expand employee permissions
AI must not give an employee information they cannot access in the source system. Document, record and tool permissions are checked on every request.
- RBAC
- SSO / IAM
- inherited source permissions
02 · Data control
Control over information sent to the model
Control information transmitted by users, agents, APIs and external models. Integrate with DLP and information protection policies, including sensitive data classification and masking.
- DLP
- data classification
- masking
03 · Logging
A complete record of AI actions
Record requests, data access, AI agent actions, tool calls and operation results. Logs support audits and investigations.
- requests
- data access
- tool calls
- results
04 · External LLM control
Centralised AI Gateway
All external model requests pass through one gateway with policies, filtering and logging. Direct employee access to public AI services can be restricted.
- AI Gateway
- routing policies
- blocking public services
05 · AI agent isolation
Minimum permissions and isolated execution
Each agent has only the permissions it needs. Potentially dangerous operations run in a sandbox, container or another isolated environment with network and file-system restrictions.
- least privilege
- sandbox / container
- network restrictions
06 · Secret protection
Secrets stay out of model context
API keys, tokens, passwords and other secrets are not included in prompts or context. Vaults, secret stores and a controlled tool gateway are used instead.
- vault / secrets storage
- tool gateway
- key rotation
07 · Prompt injection
Protection against instruction injection
A dedicated protection layer against prompt injection, indirect prompt injection, instructions embedded in documents and web pages, and unauthorised tool calls.
- input and output filtering
- separating data from instructions
- action confirmation
08 · Monitoring
Anomalous behaviour monitoring
Monitor unusual behaviour, bulk information extraction, unexpected agent actions and requests to prohibited sources. Events are sent to the customer's SIEM and SOC.
- SIEM / SOC
- anomalies
- limits and alerts
07 · Standalone service
Enterprise AI audit
We review the AI services, local models, RAG and AI agents already used by your company. The audit is a standalone project with no requirement for subsequent deployment.
The result: a clear picture of which data goes where, who can access what through AI, which risks are critical and what to address first.
01
Inventory
Identify the AI services, LLMs, agents, plugins and integrations actually used in the company, including unregistered ones.
02
Data-flow map
What is sent to the model, where it is processed and stored, and who receives the result.
03
Access analysis
Which documents, systems, databases and APIs users and AI agents can access.
04
Architecture analysis
- LLM, RAG, Vector DB
- MCP, API, RPA
- Agents, External AI
- Gateway, IAM
05
Security assessment
- access and credentials
- network and segmentation
- logging and DLP
- agent permissions and external connections
06
Attack scenarios
- prompt injection, indirect prompt injection
- leakage through context
- compromised API keys
- excessive permissions and unauthorised tool use
- access-policy bypass
07
Regulatory analysis
Identify requirements specific to the customer's infrastructure, based on the system category and the nature of its data.
08
Report
- identified risks and severity
- recommendations
- target secure architecture
- remediation roadmap
Request an AI infrastructure audit
To start, provide a list of the AI services in use and a diagram of your information systems. Audit scope and timing are agreed after a short introductory interview.
08 · Regulatory requirements
AI protection aligned with information-system requirements
Enterprise AI deployments take account of requirements applicable to the customer's information system, facility category and the nature of the information processed. Enterprise AI is not a separate regulatory object in itself: its requirements are determined by the system it becomes part of.
Depending on the project, this may include requirements concerning:
ITBP JSC carries out licensed information protection activities, including the development and manufacture of confidential information protection tools. It supports customers in preparing information systems for conformity assessment and regulator inspections, including those by FSTEC of Russia.
Scope of AI environment protection work
- security analysis
- threat model
- security control
- secure environment
- information-flow control
- data leakage prevention
- event auditing
- documentation preparation
- security testing
- preparation for conformity assessment
- preparation for regulator inspections
- operational support
The scope is agreed after assessment and depends on the class and category of the customer's information system.
09 · Project stages
From assessment to production operation
The project follows an information-system deployment process: assessment, design documentation, a pilot using customer data, security analysis and handover to operations.
01
Assessment
Business processes, systems, data, infrastructure and security requirements.
02
Use cases
Identify processes where AI delivers measurable results.
03
Architecture
LLMs, RAG, agents, integrations, computing infrastructure and security.
04
Pilot
A limited environment built around the customer's real use cases.
05
Security analysis
Access, leakage, APIs, agents, logs and external services.
06
Production deployment
Integration into existing infrastructure.
07
Operations
Monitoring, auditing, support and use-case development.
10 · Pricing
Pricing is determined after assessment
Architecture, integration scope, computing infrastructure and security requirements are defined individually for each project. After assessment, the customer receives a preliminary architecture, scope of work and cost estimate.
About us
LEGEND-A: an ITBP JSC business area
ITBP JSC is an integrated IT and security solutions provider. LEGEND-A is the company's enterprise AI business area; ITBP JSC delivers design, deployment, auditing, integration and support.
The company delivers full-cycle projects, from assessment and design to deployment, integration, protection and support for information and technical systems.
ITBP's information security expertise enables enterprise AI to be deployed within existing company infrastructure, accounting for access, data protection, information-flow control and AI operations requirements.
About ITBPITBP JSC capabilities
- Assessment
- Design
- Systems integration
- Information security
- AI deployment
- AI infrastructure protection
- Audit
- Tracking
Licensed to develop and manufacture tools for protecting confidential information. Full company, licence and registration details are available on the ITBP JSC website.
88 Sredny Prospekt, Vasilyevsky Island, St Petersburg · +7 (812) 214-25-50 · info@it-s-p.ru
Starting a project
Enterprise AI starts before model selection
First, define the business processes, data, architecture and information security requirements.
ITBP JSC can help design, deploy or audit your company's existing AI environment.