The large language model chatbot that has become popular around the world has been described as a "privacy black hole", and people have expressed concerns about the way it processes users' input data, which has even led to a brief ban in Italy.

Its creator OpenAI does not hide the fact that any input data may not be safe. In addition to being used to further train their models (which may cause their output to be exposed to others), the data will be reviewed by the manual auditor to check if it is used in compliance with regulations. Of course, any data sent to any cloud service is only as secure as the provider's security.

This means that any data entered into it should be considered public information. With that in mind, there are some things that should never be told—or any other public cloud-based chatbot. Let's look at some examples:

Illegal or immoral requests

Most AI chatbots have security measures designed to prevent them from being used for immoral purposes. If your question or request involves activity that may be illegal, you may find yourself in trouble. Things that should never be asked about public chatbots include how to commit crimes, engage in fraudulent activities, or manipulate others to take actions that can cause harm.

Many usage policies clearly state that illegal requests or attempts to use artificial intelligence for illegal activities may lead to users being reported to authorities. These laws vary from place to place. For example, China’s AI laws prohibit the use of AI to undermine state power or social stability, and the EU AI Act stipulates that “deep fake” images or videos generated by AI must be clearly marked. In the UK, the Cybersecurity Act stipulates that sharing explicit images generated by AI without consent is a criminal offence.

Entering illegal material or information requests that may endanger others is not only morally wrong, but also leads to serious legal consequences and reputational damage.

Login name and password

With the rise of autonomous AI, more and more of us will find ourselves using AI that connects and uses third-party services. To do this, they may need our login credentials; however, granting them access can be a bad idea. Once the data enters the public chatbot, it is difficult for us to control what happens later, and there have been cases where personal data entered by one user is leaked in responses to other users. Obviously, this can be a privacy nightmare, so it's best to avoid any interactions that involve giving AI username and account access to it unless you're completely sure you're using a very secure system.

Financial information

For similar reasons, it may not be a good idea to enter data such as a bank account or credit card number into a genAI chatbot. This information can only be entered into security systems for e-commerce or online banking, which have built-in security protections, such as encryption and automatic deletion after data processing. Chatbots don't have any of these security measures. In fact, once data is entered, you can’t know what will happen, and entering this highly sensitive information can put you at risk of fraud, identity theft, phishing, and ransomware attacks.

Confidential information

Everyone has a confidentiality obligation to protect sensitive information they are responsible for. Many of these obligations are automatic, such as confidentiality between professionals (such as doctors, lawyers, and accountants and their clients). But many employees also have an implicit confidentiality obligation to their employers. Sharing business documents, such as meeting minutes or transaction records, is likely to constitute sharing of trade secrets and breach of confidentiality agreements, just like the case of Samsung employees in 2023. So it’s not a good idea to stuff all this information into ChatGPT to see how tempting it can dig out, unless you’re completely sure that the information can be shared safely.

Medical information

We all know that having ChatGPT act as your doctor and diagnose medical problems can be tempting. But this should always be done with extreme caution, especially given that recent updates allow it to “remember” and even collect information from different chats to better understand users. None of these features has any privacy guarantees, so it is best to realize that we have little control over the subsequent circumstances of any information entered. Of course, this is especially important for healthcare businesses that process patient information, as they face the risk of huge fines and reputational damage.

Summarize

As with anything we put on the internet, it is best to assume that there is no guarantee that it will remain private forever. So it's better not to reveal anything you don't want the whole world to know. As chatbots and AI agents play an increasingly important role in our lives, this will become an increasingly pressing issue and educating users about risks will be a critical responsibility of any institution that provides such services. However, we should also remember that we also have personal responsibility to keep our data safe and understand how to protect the data.

The above is the detailed content of Chat-GPT Danger: 5 Things You Should Never Tell The AI Bot. For more information, please follow other related articles on the PHP Chinese website!