AI online: Assume that everything you write can become public

It is common to use artificial intelligence (AI) to get help with everyday work. Easily accessible and efficient services (chatbots) such as ChatGPT from OpenAI can effectively answer questions, compile or translate texts, transcribe audio recordings and generate images from text.

Many of these services share your information publicly on the internet in various ways. This places demands on your organisation to protect itself. One approach can be to clarify what applies in the company’s information security policy and ensure that it is followed.

• Information that is entered may be used in responses to other users. Therefore, only information classified as open should be made available to AI.
• AI is trained on information publicly available online. This also includes material protected by copyright, both within and outside Sweden and the EU. There is therefore a risk that responses may contain protected material. Sharing it further could constitute copyright infringement.
• You cannot fully rely on answers from AI. When the chatbot can’t find the answer to a question, it sometimes makes up or fabricates all or part of the answer. This is usually referred to as hallucinating. The answer can be entirely false, partly false, or even biased.
• Alongside the development of AI, methods are being developed that can reveal whether information has been generated by AI. This happens, for example, through tracking services that warn when information may have been produced with the help of AI. It is recommended to always assume that there are methods that can identify whether information originates from AI.
• Interaction with AI usually takes the form of conversations, which in many cases are saved to train the AI. In ChatGPT, for example, conversations are saved by default unless the user changes this in the settings. However, conversations are always retained for 30 days so that any offences can be investigated.

These are current issues. Last week, Microsoft temporarily blocked its own employees from using the AI service ChatGPT from OpenAI, in which Microsoft has invested heavily. This is according to CNBC. Microsoft internally stated that anyone using third-party services must exercise caution given privacy and security concerns. Microsoft later emphasised that it encourages employees to use enterprise versions of services such as Bing Chat and ChatGPT, which are said to be better protected.

With the above in mind, AI can become a valuable aid and source of inspiration for many, both in working life and in private life.

Seclight is a consultancy firm focused on digital security.

Contact: David Hässler, telephone 08-94 55 99 david.hassler@seclight.com