Universal AI Code of Conduct
A Universal AI Code of Conduct is a voluntary set of rules that companies are meant to follow when building and operating AI systems. It is not a law but a self-commitment intended to apply worldwide, often initiated by international organizations.
Programs that independently write texts, generate images, or sort job applications reach deep into people’s lives. So far, there is no uniform set of rules for such programs that applies everywhere in the world. A Universal AI Code of Conduct is an attempt to create such a set of rules. It describes what companies should and should not do when developing and offering such programs. It is signed voluntarily, so a court cannot enforce it. Well-known examples come from the United Nations, the G7 countries, and the European Union.
Why voluntary rules are needed at all
Technology develops faster than laws. In the EU, a law often takes several years from the first idea to taking effect. During that time, several new generations of AI systems come onto the market. A code of conduct, by contrast, can be negotiated in months and revised afterward. It thus fills the gap until binding law catches up.
There is also a second problem: AI knows no borders. A model is trained in the USA, operated in Ireland, and used in Brazil. National laws then only apply in fragments. A globally recognized code is meant to prevent companies from simply shifting to countries with looser rules. Experts call this shifting regulatory arbitrage.
Critics counter that voluntary rules mainly serve self-presentation. Anyone who doesn’t comply need not fear any penalty. The English technical term for this is ethics washing: companies advertise responsibility but change little. Whether a code actually works therefore depends heavily on whether anyone checks compliance.
What such a code typically contains
Almost all codes contain the same basic building blocks. First, transparency: users should be able to recognize that they are talking to a machine or seeing a generated image. Second, safety testing before release, often called red-teaming. In this process, experts deliberately try to lure the system into giving dangerous answers. Third, the protection of personal data, meaning information that can be linked to specific individuals.
On top of that come rules against discrimination. A system that learns from old application data can easily adopt its biases. A code then requires such effects to be measured and documented. Many texts also demand that, in the end, a human makes the decision when it comes to loans, grades, or medical questions.
It’s important to note the difference from an actual law. The EU’s AI Act sorts applications by risk and threatens fines in the millions for violations. A code of conduct, by contrast, works with reports, self-disclosures, and public pressure. However, some codes are meant as a preliminary stage: what proves effective may later be written into law.
The code in the news and in products
In business news, the term mostly comes up around major summit meetings. In 2023, the G7 agreed in Hiroshima on a code of conduct for advanced AI systems. The United Nations is working on similar guidelines for its member states. For investors, such reports are relevant because they show what requirements tech companies might face.
In everyday life, you encounter the consequences indirectly. The note “This image was created with AI” under a photo goes back to such transparency rules. Safety reports that providers publish for new models are also a response to this. Anyone who wants to know how seriously a company takes this can read these reports. Their scope and openness vary widely.