Carolina Principles

The Carolina Principles are a short list of guiding principles for dealing with artificial intelligence. This refers to computer programs that write texts, generate images, or suggest decisions. The list was created in 2023 at the University of North Carolina at Chapel Hill, whose nickname “Carolina” explains the name. It is not a law and is not legally binding on anyone. It is a voluntary self-commitment that a university and its members can use as guidance. The core message is: new technology may be used, but not blindly and not without responsibility.

Why a university gives itself its own rules

In late 2022, chatbots like ChatGPT came onto the market and were everywhere within weeks. Universities faced a problem: their examination regulations simply did not know these tools. An essay that is produced in ten seconds does not fit into any rule written for library work. Many institutions initially reacted with bans.

The Carolina Principles chose a different path. Instead of a ban, they rely on disclosure and reflection. Anyone who uses AI should say so and retain responsibility for the result. Behind this lies the sober assessment that bans are hardly enforceable in practice. No examiner can reliably determine whether a text was written by a human.

The principles became known beyond the university because they were formulated early and concisely. Other universities at home and abroad have used them as a template for their own sets of rules. In debates about AI in education, they therefore still appear as a point of reference today.

What the guiding principles specifically state

The most important principle concerns transparency. Anyone who uses AI for a term paper, a presentation, or an administrative decision should make this clear. This keeps it comprehensible to others how a result came about. Concealed use is considered deception, while disclosed use generally is not.

A second point is human responsibility. A chatbot occasionally invents sources, figures, or entire quotes that sound convincing yet are false. Experts call this hallucination. According to the Carolina Principles, it is always the human who submits the result who is liable, never the program. “The AI wrote that” is therefore not an excuse.

Further guiding principles concern data protection and fairness. Confidential data, such as grades or health information, does not belong in a chatbot whose operator could read the inputs. And because AI systems learn from old data, they can repeat old biases. Anyone who uses them for applications or admissions must check this. Finally, the principles demand that no one be excluded simply because they cannot afford a paid tool.

From the lecture hall to corporate policies

The principles are encountered most directly by pupils and students in guidance notes on the topic of term papers. Phrases such as “AI use is permitted but must be documented in the appendix” often trace back to this model. German universities, too, now work with very similar rules.

In business news, the term appears in connection with corporate self-commitments. Many companies have their own AI policies that follow the same structure: disclose, verify, retain responsibility, do not enter sensitive data. Experts call this kind of thing soft law, i.e., rules without the threat of penalties. They are adopted more quickly than a law and often take effect even before legislators react.

It is important to distinguish this from the EU AI Act. That is binding law with fines, whereas the Carolina Principles are not. A common misconception is also that they are a technical standard. They do not prescribe any programming, only behavior.