Shadow AI is already in your company.
AI is now part of daily work. Most of it happens outside your control, and your data goes with it.
Shadow AI
Employees paste contracts, code and customer data into personal AI accounts. The data leaves the company with no record.
No visibility, no proof
IT cannot see who used which tool or what was shared, and cannot prove to an auditor that nothing leaked.
Bans do not work
Blocking public AI pushes it onto private phones. Employees need a safe tool that is just as good.
What's included
Private Claude Deployment or another subscription?
Per-seat AI tools such as ChatGPT Business, Claude Team and Microsoft 365 Copilot use similar models. The difference is where they run, who keeps the logs and how you pay.
| Topic | Private Claude Deployment | Per-seat AI subscriptions |
|---|---|---|
| Where it runs | Your own AWS account | The vendor's cloud |
| Where requests are processed | EU regions only, recorded in every log | Depends on the plan; EU processing often only in Enterprise contracts |
| Logs of every request | In your account, masked and tamper-proof | With the vendor; full audit logs usually only in Enterprise plans |
| Control per team | Models, tools and budgets per group from your identity provider | Admin settings that vary by plan |
| Pricing | Pay for use, no seat licence | $20 to $30 per user per month, used or not |
| Policy, DPIA input, works council, training | Part of the service | Not included: you get the software |
| If you leave | Code, configuration and logs stay with you | Data and logs stay with the vendor |
You own it. We run it.
The table your CISO will ask for. You keep the account, the keys, the data and the logs. We handle the setup and the day-to-day operation, so your team just uses Claude.
- AWS account and network You
- Encryption keys You
- Region where data is processed You
- Identity provider, groups and access You
- Audit logs and activity records You
- Setup, updates, monitoring and support Sedmi odjel
- AI policy, DPIA input and staff training Together
- Training AI models on your data Never
Start small, prove the value, scale
A mid-sized company goes from the first workshop to company-wide use in about 8 to 12 weeks. We deliver a proven setup in your own AWS account: the gateway runs on ECS Fargate or EKS behind an internal load balancer, reachable only from your private network, with no internet access for the servers.
One session to map how your teams use AI today and plan a proof of concept on your own data.
Why Sedmi odjel
What IT, security and legal ask first.
Have a question that is not here? Schedule a call and bring the hardest ones.
Which Claude models can we use?
Claude Opus, Sonnet and Haiku run through EU routing on Amazon Bedrock, so requests stay in EU regions such as Frankfurt, Ireland, Paris and Stockholm. Each group gets its own list of allowed models. Models that are only offered with global routing, which can process data outside the EU, stay switched off unless you decide otherwise.
Do we need to buy Claude licences?
No. There are no seat licences. You pay AWS for actual usage, directly or through us as an AWS reseller, plus the project and, if you choose it, our managed service. Light users typically cost a few dollars a month in AI usage.
Is our data used to train AI models?
No. Amazon Bedrock does not use your prompts or answers to train models and does not share them with model providers such as Anthropic. Anthropic's terms for Bedrock also forbid training on customer content, and you own the outputs.
We already have Microsoft 365 Copilot Chat. Why add this?
Private Claude Deployment keeps every log in your own account, processes requests only in the EU, lets you decide which models, tools and budgets each team gets, and gives developers Claude Code through the same controls. Many companies use both.
How long does it take?
A proof of concept runs in two to three weeks after the workshop. A full rollout typically takes 8 to 12 weeks for a mid-sized company and four to six months for an enterprise rollout in waves.
How does this help with GDPR, the EU AI Act and works councils?
Personal data is masked before it reaches the model and processed only in the EU. We provide the AI usage policy, technical input for your DPIA, a privacy-friendly logging design that cannot be used to evaluate employees, a works council template for Germany and Austria, and the AI literacy training Article 4 of the EU AI Act requires.
What do you need from us?
An AWS account structure, or permission for us to set one up; an administrator for your identity provider; a decision on network access (VPN or protected internet access); the document sources in scope; and a sponsor, an IT contact and your data protection officer.
Who runs it after go-live?
Your IT team manages users, groups and rules through the admin interface. We can operate, monitor and update the platform as your AWS managed service provider, test new Claude models before you switch and report on usage and cost.
Sedmi Odjel Delivers AI-Powered Legal Research Platform on AWS
Sedmi odjel built a generative AI solution enabling a Croatian law firm to transform legal research through intelligent court decision analysis.
Read case studyTell us about your company and we will come back with a plan for your first step.