Artificial Intelligence, strictly speaking, is not intelligent. Instead, it generates excellent results based on pattern recognition learned with massive data sets. Modern publicly available AIs also learn from the information provided to them. Hence, all information sent to an AI becomes practically public afterward. You should not send any confidential information to an AI. For those operated outside of Europe, the privacy issue (already known from other cloud services) is an additional concern.

AIs are software, which means they can be hacked like any other software. Certainly, hosting companies implement appropriate security measures, but a successful attack cannot be entirely ruled out. Therefore, trust an AI similar to any other cloud service; if you notice suspicious behavior or unusual communication, end your interaction with the AI.

AIs pose yet another hacking risk: manipulation of their "decision-making." This occurs through targeted changes to the data used for their learning (which may include user-provided data). As a user, it is almost impossible to detect such manipulation. Hence, do not rely entirely on AI results—they may be manipulated or simply incorrect.

However, AIs are not solely a risk. They can be a powerful ally in ensuring cybersecurity—for instance, by supporting the creation of tailored security concepts and identifying suitable measures to mitigate significant risks.