Personal Liability for 30,000 German Firms Drives Shift to On-Premise AI as Microsoft Unveils Cost-Cutting Tools
15.06.2026 - 03:15:40 | boerse-global.de
June brought a double dose of AI announcements from Microsoft, but the real story for German businesses may be the regulatory pressure forcing many to rethink where their data lives. The NIS-2 directive, which applies to an estimated 30,000 companies, makes company leaders personally liable for implementing security strategies — a threat that is pushing some toward on-premise AI solutions over cloud-based services.
On June 11, Microsoft introduced “DigitalMe,” a personal AI twin designed to sit in on meetings and handle moderation and minutes automatically. In tests, the system correctly answered 90 percent of more than 150 technical questions within an hour. Each session saves between 60 and 90 minutes of moderator time. New integrations with platforms like Miro and GitHub allow results to flow directly into existing workflows.
Just three days later, the company announced the MAI model family. The flagship model, “MAI-Thinking-1,” packs 35 billion parameters and promises to cut the operating costs of AI applications by up to 90 percent. Experts stress that these tools are not replacements for human workers, but rather relieve them of routine tasks such as automated accounting or process control. Industry solutions for bookkeeping already report throughput time reductions of up to 80 percent, using pre-trained language models.
Yet the same tools must comply with strict regulations. June’s professional conferences warned of the risks ChatGPT or Microsoft Copilot pose for small and midsize enterprises. The focus is on the EU AI Act, the GDPR, and the NIS-2 directive. Specialized AI assistants are already helping companies create records of processing activities or data protection impact assessments — cutting the time needed by 60 to 75 percent. But the final legal review remains a human responsibility.
Given the personal liability under NIS-2, some vendors are turning to on-premise installations: powerful language models that run on a company’s own servers without sending data to the cloud. In June alone, over 200 security updates were released for common office software. And the support deadlines for older SharePoint versions — 2016 and 2019 — expire this summer, adding another reason for firms to review their security architecture.
