BSI situation report: what it means for AI in your company

The BSI rates Germany's cyber situation as tense to critical and names the rapid development of AI as one reason. Companies that use AI with confidential data can turn this into a short list of checks on access, logging, suppliers and emergency planning.

Sandra Leistner4 min read

Illustration: A wooden watchtower with a bell and a latched crate holding a lantern stand for vigilance and protection.

What the BSI found in its annual situation report

In November 2026 the Federal Office for Information Security (BSI) published its annual report and described the situation in cyberspace as tense to critical. It names two reasons: the geopolitical situation and the fast development of artificial intelligence. For the first time the BSI also presents a Cyber Index, which condenses risk factors into one figure. The index averages 170 points, 50 percent more than in the previous reporting period.

The report counts 160 newly disclosed vulnerabilities per day, 35 percent more than a year earlier. The time between a vulnerability becoming known and its first known exploitation fell from 53 days to 25 days, and to only 8 days in the second quarter of 2026. Attackers often know a flaw before it is public or before a patch exists.

The BSI repeats that small and medium-sized companies and municipalities are particularly vulnerable targets. It also finds that fewer than half of users of digital applications apply basic security measures. In Germany the Mittelstand, meaning the owner-led small and medium-sized firms, is therefore addressed directly.

Why AI changes how you review your IT security

The BSI president, Claudia Plattner, said that high-performance AI enables attacks in volumes and at speeds not seen before. Interior Minister Alexander Dobrindt added that more use of AI means more points of entry. For a company this has two sides. Attackers become faster, and your own AI tools add systems that touch sensitive content.

An AI assistant that reads contracts, mails or drawings concentrates access to much of your knowledge in one place. A weak account, an over-broad permission or an unvetted plug-in can then expose far more than a single file. AI therefore belongs in the regular security review, and a separate pilot folder keeps it out of sight.

Personal data in prompts and documents also brings the GDPR into play. Professionals such as lawyers and tax advisers must consider section 203 of the German Criminal Code, which protects client secrets. Security and data protection questions overlap here, and the same inventory serves both.

Which checks apply to AI with confidential data

The report does not list AI-specific controls, so the following points are common practice derived from its themes. Work through them with IT, the data protection officer and, where one exists, the Betriebsrat (works council), since logging of user activity touches employee data.

  • Access: the AI should see only what the asking person may see. Check that it inherits existing permissions, that roles are separated and that former staff and test accounts are removed.
  • Logging: record who asked what and which sources were used, keep the logs protected, and define who reviews them and how long they are kept.
  • Suppliers: list every model provider and hosting location involved, ask where prompts are processed and whether they are stored, and review the contracts accordingly.
  • Write access: let AI agents prepare changes in mail, tickets or files, but require a named person to approve anything that writes to a system.
  • Authentication: use multi-factor login for every AI tool, since the BSI notes that fewer than half of users apply basic measures.

Patching and emergency planning on a short deadline

With an average of 8 days in the second quarter between disclosure and first exploitation, a monthly patch cycle leaves a gap. Many attacks start before a patch exists, so patching alone is not enough. Systems that face the internet, including AI gateways and their connectors, deserve the shortest cycle and a named owner.

Emergency planning should cover the AI components explicitly. Ask which systems you can switch off at short notice, how an AI service can be cut from the document sources it reads, and how work continues without it. Keep backups offline, test restoring them, and rehearse who decides and who informs customers and authorities.

Reduce what can be reached in the first place. A model running in your own network or hosted in Germany shortens the list of parties you must assess, while external services need a clear record of what data they receive. Review this list at least once a year and after every change of provider.

Working with AI and company documents securely

The checks above apply to any AI that works with company documents, and souveraen.ai is built for that task. The platform answers from connected sources such as SharePoint, file servers, Nextcloud or mailboxes, read-only and with existing permissions, and it cites document, version and page for each answer. Agents prepare tasks, and anything that writes to a system waits for approval. Administration offers roles, workspaces and a usage overview.

It runs in two modes with the same software: On Demand, hosted in Germany, or on-premise in your own network as Private Spark. The model catalogue includes local models, models hosted in Germany and external EU models with zero data retention. Holders of professional secrets can sign a confidentiality undertaking under section 203 StGB. More detail is on the security page.

Sources

More articles

All articles

How safe are AI API routes via resellers and routers?

Security firm Dragos sent about two million requests with fake credentials to AI providers and found compromised services and resellers reading customer prompts. Companies that use AI through third parties need to know who sits on the path of their data.

Data protection4 min

Run Text Generation Inference with Docker on Your Server

Text Generation Inference (TGI) serves open language models from a single Docker container on your own hardware. This tutorial starts it, sends requests over HTTP and keeps prompts inside your network.

Tutorial4 min

Cloud outage liability: who answers when Microsoft fails

On 12 November 2026 a software fault took Exchange Online and Outlook offline for about 36 hours. A legal analysis published shortly afterwards argues that responsibility for such outages often lies with the operator rather than the provider.

Regulation4 min

Test it with your own documents.

Free for 3 days, then from €24.90 per user per month. Cancellable monthly.

Free for 3 days, then from €24.90 per user per month.Free for 3 days
Try free for 3 days