September 16, 2026, 3:47 pm | Read time: 3 minutes
Artificial intelligence is now taking on many tasks. At the same time, questions are growing about how reliable the results of such applications really are. This is precisely where a new certification program by the TÜV Association aims to step in. In the future, a certification mark could indicate whether an AI has been independently tested for quality and safety.
The TÜV Association is planning three different certifications for this purpose. The rollout is scheduled to begin gradually after a pilot phase with partners at the end of 2026. The individual offerings differ in terms of testing procedures, scope, and depth. However, the certificates will not be mandatory and will not replace legally required inspections. For certain high-risk applications, additional requirements already apply. These include systems for job application processes, credit assessments, or law enforcement. Stricter regulations will apply to them starting at the end of 2027.
Three Certifications for Different AI Systems
The first certification is aimed at AI systems that are not classified as high-risk applications under the European AI Act. Experts will check basic requirements for quality and safety. The goal of the assessment is to determine whether the respective systems meet the requirements intended for their use.
The second variant, on the other hand, is open to all AI systems. Inspectors analyze potential risks, review documents, and check implementation directly with the manufacturer. The third certification goes a step further. Here, independent experts additionally test selected features of an application. This is intended to allow for a more precise assessment of a system’s performance and reliability.
Security Incidents Bring the Issue into Focus
The announcement comes against the backdrop of recent incidents in the AI industry. At OpenAI, models ended up on the internet during a security test. They operated with limited protective measures and accessed external systems. Anthropic also recorded four similar cases. According to the company, a misconfiguration in the test environment led to the incidents.
Also of interest: AI develops its own language–researchers don’t understand it
At the same time, figures from the representative TÜV AI Study 2025 show that many people view the potential risks of artificial intelligence critically. According to the study, 79 percent of Germans fear incalculable dangers from the technology. Even among users of AI systems, just over half of respondents trust the generated answers and results. According to the TÜV Association, trust therefore plays a central role in the further spread of the technology.