Artificial Intelligence – Productive and Without Compromise

Artificial intelligence promises enormous productivity gains, but it also raises many questions about data protection and responsible use. In the right setting, both can be achieved. In this article—and explored in greater depth in the “AI Workshop”—you’ll learn how to use AI safely and effectively with minimal effort.

Artificial intelligence (AI) has transformed the day-to-day work of many organizations in a short period of time. Texts are produced more quickly, analyses can be automated, and ideas are generated at the push of a button. What sounded like science fiction just a short time ago is now part of everyday life. With minimal effort and the necessary basic knowledge, an organization can set up a secure AI system that effectively supports its day-to-day operations (see box below: “AI Workshop”).

Along with this new productivity comes a new sense of uncertainty. What actually happens to the data we enter into AI systems? Who stores it? And could our content potentially be used to train third-party models?

The Dilemma: Safety or Efficiency?

The issue of security is particularly crucial for nonprofits and funding organizations. This is because they often work with sensitive information: project dossiers, funding decisions, strategies, or personal data. The challenge, therefore, is not just to use AI—but to use it responsibly and securely.

Data protection is crucial. Many of the popular AI tools are based on cloud infrastructures provided by U.S. companies. Often, it is not clear what happens to the data that is entered. It is not only the handling of personal data—as regulated by data protection laws—that is sensitive. Almost as critical is the handling of trade secrets and internal strategies, or of intellectual property such as concepts or texts—especially when the latter originate from third parties.

It’s an open secret that organizations—especially those that don’t have answers to these questions or that respond by banning all use of AI—are making a huge mistake. If the use of AI tools is prohibited, employees will simply turn to the free versions of ChatGPT and similar services. And it is precisely the free versions that offer absolutely no data protection. On the contrary: here, users “pay” with their data.

PeakPrivacy: Data Sovereignty Is a Priority

This is a problem for organizations that handle sensitive data.

The responsible use of artificial intelligence was therefore also the focus of the “AI Learning Journey,” conducted in collaboration with SwissFoundations, which Spheriq launched in late 2023. It became clear that what matters is not only what AI can do, but also the conditions under which it is used. As a result of this learning journey, a framework for the responsible use of AI was developed: PeakPrivacy.

PeakPrivacy focuses on organizations’ data sovereignty and combines conceptual guidelines with a technical solution. The underlying AI models are run on the organization’s own servers, ensuring that sensitive data never leaves the organization’s infrastructure. Three principles are central to this approach:

• First, the AI runs on our own infrastructure.

• Second, no data is shared with external platform providers.

• Third, users’ prompts and content are not used to train the models.

This makes it possible, for the first time, to consistently combine productivity and data protection.

What AI Can Really Do—and What It Can’t

To use AI effectively, it helps to start with a realistic understanding of its capabilities. Essentially, artificial intelligence refers to a machine’s ability to perform tasks that normally require human thinking—such as writing texts, summarizing information, or recognizing patterns.

So-called Large Language Models (LLMs)—the core of today’s AI—are particularly effective at tasks such as generating text, summarizing content, answering questions, classifying or translating text, and formulating suggestions, ideas, or recommendations.

At the same time, these systems have clear limitations. For example, they can internalize biases from training data. It is not uncommon for complex relationships to be misinterpreted—but presented with such superficial clarity and conviction that the underlying error is not easily detected. In other words: information is presented convincingly, even though it is incorrect. Another general weakness is stability: results are difficult to reproduce. But if the recommendation turns out differently every time the process is repeated, trust in the technology quickly—and rightly—wanes.

Context is what makes the tool powerful

That is why AI is less an automated form of intelligence and more a potentially powerful tool. Philosopher and AI expert Luciano Floridi goes so far as to reinterpret the acronym AI as “Agency, not Intelligence.” Here, the “A” stands for a broad capacity for digital agency. This shifts the emphasis significantly: Artificial Intelligence is a tool that provides support but should not make responsible decisions on its own.

Since the conclusion of the AI learning journey, Spheriq has not been idle either. Based on the principles developed, Spheriq AI has since emerged as a new dimension of AI support in the nonprofit sector. The idea behind it is simple: AI is particularly effective when it not only draws on general knowledge about the world but can also work with a specific dataset.

Spheriq AI therefore combines two elements: first, a powerful AI model architecture, and second, Spheriq’s specific knowledge base. With agent-based tooling, various AI components can work together—much like a team of specialized assistants. Spheriq’s specific knowledge base and core guiding concepts—such as “Fit,” “Hope,” and “Support Logic” —create the ideal context.

This combination opens up new possibilities. For example, AI assistants for grant applications help review proposals. Intelligent research in the foundation sector quickly and accurately reveals where an application is worth pursuing and where it is not. Spheriq AI also provides fast and reliable support for portfolio analysis or due diligence when evaluating potential partnerships. And the range of possibilities is far from limited to these examples.

Audit-proof rather than selective

This means that Spheriq AI’s use is also in line with the legal requirements for responsible AI use by grant-making foundations. In a recent essay on the use of AI by foundations, Thomas Sprecher states that AI must be used for specific purposes, in a controlled manner, and with clear human accountability. AI processes relevant to decision-making must remain documented, traceable, and verifiable.

This applies in particular to the preliminary review of applications. The spokesperson explicitly identifies this as a key area of application in grant management. Responsibility for the funding decision remains with the relevant bodies, but artificial intelligence may be used to structure, summarize, and prepare the basis for decisions—provided that this use of AI is documented and remains verifiable.

It is not feasible to ensure this level of auditability if inputs and outputs for general AI subscriptions such as ChatGPT must be recorded manually. Rather, what is required is an automatic, comprehensive record of AI usage: data sources used, prompts, outputs, and intermediate steps relevant to decision-making must be systematically documented. This is precisely where a key difference lies between occasional AI use and an audit-proof AI infrastructure suitable for grant-making foundations, such as Spheriq AI.

Setting Up AI for Everyday Life

With Spheriq AI, nonprofit organizations now have access to a secure AI environment that leaves nothing to be desired in terms of data sovereignty, security, and documentation. The direct connection to the nonprofit sector’s specific database ensures reliable and accurate results.

However, Spheriq AI only becomes truly effective once the system is tailored to an organization’s specific context—through predefined funding logic and a complete, transparent organizational profile. After all, not every AI application automatically adds value.

Artificial intelligence is no longer just a buzzword—it’s a reality. The challenge isn’t to use as much AI as possible. Rather, it’s about using it wisely, safely, and in a targeted manner.