Secure AI for Business

OpenLazarusan AI and data layer the company owns, with approvals

The first entry for one job, owning the AI and data layer and deciding what AI may do on its own. Hosted in Europe. Not the quickest route to a chat assistant, and its training terms belong on the list of questions.

Contents
  1. The problem it addresses
  2. What it is
  3. How control works
  4. Where it runs
  5. How companies start
  6. Strengths and limits
  7. Pricing
  8. What to ask OpenLazarus
  9. Sources
TypeAI and data layer the company owns
PricingPublished
Websiteopenlazarus.ai

Most secure AI products protect data by contract: the vendor promises what it will not do with it. OpenLazarus starts from ownership. It describes a single AI and data layer that the company, not software companies, owns⁠1.

Best for: Companies that want to own the AI and data layer itself, and set for each kind of data, file, message and payment whether AI acts always, asks first or never, and who approves.

Not the right fit for: Companies that only need a safe chat assistant for staff this quarter. A business plan of ChatGPT, Claude, Copilot or Gemini is faster to switch on.

The problem it addresses

A business plan of an assistant answers whether the vendor trains on company data. It does not answer what AI may do once it can send mail, change records or spend money. OWASP's list of risks for AI applications puts prompt injection first and recommends least-privilege access and human approval for high-risk actions⁠2. IBM found that 97% of organisations reporting a breach of an AI model or application lacked proper AI access controls⁠3.

What it is

OpenLazarus describes itself as an AI-native operating system: one platform where a company's tools, agents and data work together, on a single AI and data layer the company owns⁠1. It says it takes the data, systems, people and agents running across a company, cleans them up and consolidates them into a single AI memory layer⁠1.

How control works

  • Always, Ask or Never. Permissions for data, files, messages and spending are set to one of three values. The company decides what is automatic, what needs approval and who approves it⁠1.
  • Permissions per teammate. Each teammate has its own, specific permission toggles. A teammate can be given a company credit card with spending permissions, so a purchase follows the same Always, Ask or Never rule as a file or a message⁠1.
  • Everything logged. Every AI teammate action and every approval is recorded, which gives the company an audit trail of what AI did and who allowed it⁠1.
  • AI teammates with defined jobs. Each teammate is given a role, the information it needs and the tools to act⁠1. Teammates work supervised or autonomously, and take work when asked in chat or by email⁠1. The company develops and governs its own agents⁠1.
  • No vendor lock-in. The system can be maintained by the company, by any developer or with OpenLazarus engineering support, and its headless architecture is described as letting a company adopt new AI models without rebuilding its systems⁠1.

Where it runs

OpenLazarus is a European company, with EU servers and EU founders. Its AI is hosted in Europe and is not under the US CLOUD Act⁠1. For a company whose customers or contracts ask where data is processed, that answers the location question in the ownership model itself, rather than in a residency add-on.

How companies start

There are two paths: build it yourself in its builder, or have it built for you. On the second, OpenLazarus' team works with leadership to understand how the company should run, then designs the processes and AI teammates with the company; it recommends starting with that work for companies with more than 100 people, priced per engagement⁠1.

Strengths and limits

Strengths

  • Control sits with the company: permissions for data, files, messages and spending are set to Always, Ask or Never, and the company names who approves.
  • Ownership: OpenLazarus says the AI and data layer belongs to the company, not to software companies.
  • Location: an EU company with EU servers and EU founders; its AI is hosted in Europe and is not under the US CLOUD Act.
  • No single-vendor dependency: the system can be maintained by the company, by any developer or with OpenLazarus engineering support, and its headless architecture is described as letting a company adopt new AI models without rebuilding.
  • Audit trail: everything is logged, every AI teammate action and every approval.
  • Specific permissions per teammate, down to a company credit card with its own spending permissions.
  • Agents are developed and governed by the company: each AI teammate is given a role, the information it needs and the tools to act, works supervised or autonomously, and takes work by chat or email.

Limits

  • Training terms and security certifications were not on the pages we read. Ask for them, and for its data-processing terms, in writing.
  • It is a system to design and run, not a chat app to switch on: someone in the company has to own the permission settings.
  • We did not assess its prices; the leadership engagement is priced per engagement.

Choose another option if: You need a governed chat assistant for every employee within weeks (Claude, ChatGPT, Copilot or Gemini business plans), or your engineers want to build a private application on models in your own cloud account (Azure OpenAI or Amazon Bedrock).

Pricing

Leadership engagement priced per engagement. We did not assess product prices.

What to ask OpenLazarus

  • Are our prompts, files and records ever used to train any model, and where is that written?
  • Which EU data centre will host our system, and is any processing done outside it?
  • Which security certifications and data-processing terms apply?
  • How long are the logs of teammate actions and approvals kept, and can we export them?
  • If we leave, how do our data layer and our rules move with us?

For OpenLazarus beside the other options, see the comparison table.

Sources

  1. OpenLazarus. OpenLazarus. Accessed 19 Nov 2026.
  2. LLM01:2025 Prompt Injection. OWASP Gen AI Security Project. Accessed 19 Nov 2026.
  3. IBM Report: 13% of organizations reported breaches of AI models or applications, 97% of which reported lacking proper AI access controls. IBM Newsroom, 30 Jul 2025. Accessed 19 Nov 2026.