ⓘ This page has been translated using artificial intelligence.
Artificial intelligence: opportunities, risks and applications
AI-generated content has become an integral part of everyday life. The colleague who structures his annual report with ChatGPT. The student who has a graphic created for her presentation. Or the promotional video for an SME that was created without a single day of filming. Artificial intelligence is changing the way we work, learn and communicate. And faster than we would have expected. But what is behind the technology? What can it really do, and where are its limits? And what does that mean for us?
Go directly to topic
Share this page
You will find these topics on this page:
- What does artificial intelligence mean, how does AI work, and where do we encounter it?
- What are the best-known generative image and text AIs?
- What are the opportunities and dangers of AI?
- And what issues should parents discuss with their children?
You will find the following topics on this page
Multimodal AI goes beyond text and images.
Topic
How does generative AI work?
Generative artificial intelligence is based on algorithms that it uses to solve specific tasks. There are several approaches, such as machine learning and deep learning. This may sound complicated, but basically it always involves artificial intelligence attempting to mimic the performance of our brain.
Machine learning, deep learning, computer vision or natural language processing – what do these terms mean?
Machine learning
In machine learning, AI attempts to identify specific patterns in huge amounts of data. By analysing this data, it learns to always find the most likely answer to a specific question or task. However, the most likely solution is not necessarily the correct one. It is always possible that generic AI will invent an answer, i.e. hallucinate. That is why it is important to cross-check important information with other sources.
Deep learning
As in the human brain, many artificial brain cells (known as neural networks) send signals to each other during deep learning in AI. The AI builds up its knowledge by linking these artificial brain cells. We humans also learn by linking our brain cells together.
Natural language processing
One early application of AI was natural language processing. Here, AI (for example in voice assistants or chatbots) analyses and processes human language and learns to respond intelligently to questions.
Diffusion models and autoregressive models
Image AIs work fundamentally differently from language models. For a long time, so-called diffusion models dominated: they start with random image noise. Imagine a grey, grainy surface. This is gradually refined until an image emerges that matches the prompt. Models such as Midjourney or Stable Diffusion work according to this principle and are particularly strong at creating atmospheric, artistically designed images.
Newer models such as GPT Image or Nano Banana Pro, on the other hand, are autoregressive models: they do not generate an image by refining noise, but build it up step by step (similar to how a language model builds up a text). Since these models understand text and images together, they follow instructions more precisely and are easier to adjust in conversation: ‘Make the background darker’ or ‘Add a person’. An autoregressive model understands this directly.
Transformer architecture and large language models (LLMs)
Modern language models such as ChatGPT, Gemini and Claude are based on what is known as transformer architecture, a breakthrough from 2017 that fundamentally changed AI development. A transformer does not simply understand individual words, but their meaning in context. It recognises that ‘bank’ in the sentence ‘I am sitting on the bank’ means something different than in ‘I am going to the bank’. This understanding of context is what makes modern language models so powerful.
Building on this architecture, large language models (LLMs) were developed. These are models that have been trained with enormous amounts of text and have thus learned to recognise and predict linguistic patterns. Put simply, an LLM calculates which word is most likely to fit next in each response and does so step by step until a complete text has been created.
Machine learning, deep learning, language processing, transformer architecture and image generation together form the basis for what you experience today as generative AI in everyday life: so-called multimodal systems that can read, write, analyse and generate simultaneously.
Topic
What generative AI is available for images,
text and video?
Since the first version of ChatGPT was released in 2022, the range of AI software on the market has virtually exploded. Due to high demand, the range has been developing rapidly ever since. Here you will find an overview of the most popular models currently available.
Language models
ChatGPT (OpenAI)
ChatGPT(opens in new tab) is a multimodal model that originated as a language model from OpenAI. Its main task is to understand natural language and respond to text-based queries. This also includes response formats such as tables, code or mathematical forms. Even diagrams are possible.
ChatGPT as a service offers different models (including GPT-5.2, GPT-4.1 and gpt-oss), which differ in their functions and strengths. Since December 2025, the image model GPT Image has been available in ChatGPT for creating images. Previous models such as GPT-4o and DALL·E have been replaced.
DeepSeek
DeepSeek(opens in new tab) is a language model from China that caused a stir in early 2025: it achieved top performance at a fraction of the usual development costs.
The DeepSeek R1 model (released in January 2025) was developed for complex thinking tasks, specialising in mathematical problem solving, challenging programming tasks and logical conclusions. For the first time, it was possible to follow the AI's thought process, as the AI reveals its intermediate steps, reflects on them and corrects them independently.
This step-by-step method (chain of thought) leads to precise results and has now become standard in many language models.
The all-round DeepSeek V3 model for everyday tasks such as text creation, conversations and general questions was introduced in December 2024. DeepSeek V3.2 is currently in operation (as of February 2026). As an open-weights model (MIT licence), DeepSeek can be installed and used via a website and smartphone app as well as via a local installation (e.g. Hugging Face).
Google Gemini (Google AI, DeepMind)
Google Gemini(opens in new tab) is Google's multimodal AI model. Originally released in March 2023 under the name ‘Google Bard’, the model is now a direct competitor to ChatGPT after a bumpy start and rebranding to ‘Google Gemini’.
Gemini processes not only text, but also images, audio, video and code. This allows you to analyse photos, summarise complex documents or solve code problems, for example.
Currently (February 2026), Google offers different model families: Gemini Flash for fast tasks, Gemini Pro for demanding queries and Gemini DeepThink for complex thinking tasks. For image generation, Google offers Nano Banana (Gemini 2.5 Flash Image) and Nano Banana Pro (Gemini 3 Pro Image). And Google is no slouch when it comes to video generation either: with Veo 3 (May 2025), the company enables all users with a Google One AI Premium Plan to generate videos including synchronised audio: 8-second video clips in 4K quality can be created from text descriptions or images.
Gemini can also be directly integrated into Google Workspace apps (Docs, Gmail, Sheets, etc.) with the AI Premium subscription. Without a subscription, Gemini can be used as a chatbot free of charge, but with limited functionality.
Meta Llama / Meta AI (Meta)
Llama(opens in new tab) (Large Language Model Meta AI) is Meta's AI model family, which has been under continuous development since 2023. The current generation, Llama 4 (as of February 2026), is multimodal (text and image) and, like DeepSeek, uses a mixture-of-experts architecture. This means that only those parts that are needed for a specific task are activated. This makes the AI faster and more resource-efficient.
The Llama models are openly available and are primarily aimed at developers, researchers and companies.
Since March 2025, Meta AI, Llama's chatbot, has also been available in Switzerland. You can find it on WhatsApp, Instagram, Facebook and Messenger: simply search for the blue circle icon or type @MetaAI in a chat. For data protection reasons, the European version has limited functions compared to the US version.
Claude (Anthropic)
Claude is Anthropic's AI model family, which is particularly known for its safety focus. The current generation, Claude 4.5 (as of February 2026), offers three variants: Haiku (fast and inexpensive), Sonnet (balanced) and Opus (highest performance). Opus 4.6 followed in February 2026 with enhanced capabilities.
Claude excels at creative writing, code generation, analytical reports and structured documents. With the Extended Thinking feature, the model shows its thought process and reflects on intermediate steps to provide more precise answers. The new Agent Teams feature (available from Opus 4.6) makes it possible to divide complex tasks among multiple AI agents working in parallel.
Anthropic attaches great importance to data protection and develops Claude according to ethical principles.
Mistral AI (frontier AI)
Mistral AI(opens in new tab) is considered a leading European company for AI language models. The French start-up was founded in April 2023 by former Google DeepMind and Meta researchers. Since then, Mistreal has deliberately positioned itself as a European, privacy-friendly alternative to US chatbots with its open-weight models.
The current Mistral Large 3 model (as of February 2026) and the Ministral 3 family offer multimodal capabilities (text, image and audio) and support dozens of languages. The chatbot ‘Le Chat’ was launched in February 2024 and offers features such as Voice Mode for speech recognition, Deep Research for structured research and Think Mode for complex thinking tasks.
Mistral AI attaches great importance to GDPR compliance and offers both free and enterprise versions for businesses.
Microsoft 365 Copilot (Microsoft)
Microsoft 365 Copilot(opens in new tab) is Microsoft's AI assistant and is integrated directly into all Microsoft apps (Teams, Word, Excel, PowerPoint, Outlook). The assistant is based on OpenAI technology (currently GPT-5.2, as of February 2026).
Since 2026, Copilot has even been operating in Agent Mode as an autonomous agent that can perform multi-step tasks independently and across multiple apps. Instead of just responding to commands, Copilot works iteratively with you, creating drafts, refining them and adjusting documents step by step. Thinking Mode activates advanced thinking skills for complex analyses. Voice Mode enables voice control. Implicit Grounding automatically uses the context of open emails or highlighted text to provide more accurate responses.
This makes Copilot a great sparring partner for everyday tasks such as creating Excel formulas, designing presentations with information from Word documents, summarising email conversations or collecting meeting notes. Copilot respects your permissions and security settings.
myAI (Swisscom)
myAI(opens in new tab) is Swisscom's AI assistant for everyday life in Switzerland. It is based on Anthropic's Claude language model and is specifically tailored to Switzerland.
myAI helps you, for example, to write, translate and summarise texts, analyses uploaded documents (PDF, Word, Excel) and generates images. The integrated web search provides up-to-date information, with Swiss sources being prioritised.
Swiss services are also integrated, such as SBB timetables, weather forecasts from MeteoSchweiz, Swiss spelling and, because myAI is a Swiss product, the assistant also understands and speaks Swiss German.
The data remains under Swisscom's complete control. It is not passed on to third parties or used for AI training. You can delete all your conversations at any time. myAI is available to all Swiss residents with a Swisscom login.
Apertus
Apertus(opens in new tab) is Switzerland's first large language model. It was developed by EPFL, ETH Zurich and the Swiss National Supercomputing Centre (CSCS) and released in September 2025.
The name says it all: apertus is Latin and means open. A fitting choice, as the language model is completely open, meaning that anyone can see what data the AI was trained on, what method was used and how it works technically. This makes Apertus the first large language model to comply with the transparency requirements of the EU AI Act and it was developed in accordance with Swiss data protection law and copyright law.
Apertus is available for download on Hugging Face(opens in new tab).
PulicAI(opens in new tab) serves as a free chat and API demo interface.
Perplexity
Perplexity(opens in new tab) is an AI-powered search engine that, unlike traditional chatbots, is based on real-time web searches rather than stored (trained) knowledge. Perplexity specialises in fact-based information retrieval and is particularly valued by students and researchers.
Perplexity is a multi-model platform and incorporates different models such as GPT-5, Claude, Gemini, etc., depending on the version. With Deep Research, Perplexity creates comprehensive reports, cites sources cleanly and exports them as PDFs. So-called Focus Modes tailor the search results specifically.
Perplexity is free to use. The Pro version offers advanced features and model selection.
Image AIs
GPT Image (OpenAI)
GPT Image(opens in new tab) is OpenAI's current model family for image generation. GPT Image is the successor to DALL·E, which still functioned as a separate diffusion model. GPT Image is now natively integrated into the language model. This means that text and images are created in the same neural network. This is convenient for you because you can continue working with the image directly in the chat, formulate adjustments in natural language, and the model understands the entire context of your conversation.
With GPT Image, you can also edit images. The model makes very precise changes to what you request (e.g. image content) while keeping exposure, composition and facial features consistent.
GPT Image is available in ChatGPT, but it can also be accessed via Microsoft Copilot or Apple Intelligence.
Midjourney
Since its launch in 2022, Midjourney(opens in new tab) has been considered by many to be the gold standard in AI image generation. The model is known for its exceptional stylistic depth and artistic quality. This makes the model ideal for you if you want to create something visually expressive and unique rather than just a functional image.
The current version 7 (as of February 2026) has been completely rebuilt from the ground up, with more precision in the details, more realistic textures and lighting. You can also formulate your image requests via voice input. Midjourney adapts to your visual taste through personal style calibration.
Since June 2025, video generation has also been possible: you can animate an existing image into short video clips. Midjourney is accessible via the web interface and Discord.
Nano Banana Pro (Google)
Nano Banana Pro(opens in new tab) is Google's new image generation and editing model based on Gemini 3 Pro. It stands out from other models such as GPT Image or Midjourney with its photorealistic, natural-looking results. You can also influence the camera angle, lighting, depth of field and colouring in a very targeted manner. Images can be generated in up to 4K resolution, and text within images can be correctly displayed in different languages, fonts and styles.
Nano Banana Pro is available in the Gemini app, Google Slides, NotebookLM and other Google products. The image model can be used free of charge with a limited quota.
Firefly (Adobe)
Adobe Firefly(opens in new tab) is Adobe's AI image generator and is integrated directly into Photoshop, Illustrator and Premiere. All Firefly models have been trained exclusively with licensed content, meaning that images generated with them can be used commercially without hesitation. Firefly also functions as a model hub, integrating partner models such as GPT Image and Nano Banana Pro.
Adobe Firefly is available via your browser and Adobe Creative Cloud. Limited generations are possible with a free Adobe account.
Canva AI (Dream Lab)
Canva(opens in new tab) is the most popular design platform for beginners and students. Dream Lab now features powerful image AI. You can generate your images directly in your flyer, presentation or social media post without having to leave the platform. Since the acquisition of Leonardo.ai, Dream Lab delivers significantly better results than before.
Canva AI is accessible via browser and app. With a free account, only limited generations are possible.
Open-source models
In addition to commercial image AI, there is a vibrant world of open-source models, including Stable Diffusion, FLUX and HunyuanImage. They are all free to use, can be freely customised and run on your own computer. This makes them particularly interesting for tech-savvy users and data protection-conscious institutions that do not want to send data to external servers.
Setting up and maintaining such AI image models requires technical expertise and powerful hardware. Those who opt for open-source models benefit from maximum control and an active community that continuously publishes new models, fine-tuning and enhancements.
Video generators
Sora (OpenAI)
With Sora 2(opens in new tab), OpenAI not only released an improved video generation model in September 2025, but also a standalone social app called ‘Sora by OpenAI’. Similar to TikTok, its feed shows short, AI-generated videos that you can create, remix and share via text prompts.
Sora 2 generates videos with synchronised dialogue and sound effects and understands how people or objects behave and move in space. With the ‘Characters’ feature, you can insert yourself or friends (caution: data protection) directly into any scene after a one-time video recording.
The app is available in selected countries for iOS and Android (Switzerland is not yet included). If you want to continue accessing the image model without a social app, you can do so via sora.com or ChatGPT.
Veo (Google)
Veo(opens in new tab) is Google DeepMind's video generation model and can create videos from text or images. The model is characterised by particularly realistic motion physics and precise camera control.
Since version 3, Veo can also generate audio directly with the video. This means you can describe sound effects, ambient noise and even dialogue in the prompt, and Veo will generate a synchronised soundtrack that matches the image. Since October 2025, Veo 3.1 has further improved lip movements and speech, allowing characters to talk to each other.
With a Google AI Pro subscription, you can use Veo 3.1 Fast in the Gemini app; Google AI Ultra is required for full access. Developers can access Veo via API or Vertex AI.
Runway
Runway(opens in new tab) is one of the pioneers of AI video generation and its models are aimed at professional filmmakers, content creators and studios. The Gen-4.5 model (since December 2025) supports the familiar text-to-video, image-to-video and video-to-video input modes.
Gen-4.5 is even better than its predecessors in terms of physical accuracy: people and objects move with realistic weight and momentum, and liquids flow naturally. The result is footage that looks deceptively real. Complex scenes with multiple elements, precise composition and expressive characters that remain consistent in their gestures and facial expressions across multiple shots are the strengths of the model.
Runway can be used via the web and app. Gen-4.5 is also integrated into Adobe Firefly, which means the model can also be used in professional editing tools such as Adobe Premiere and Adobe After Effects.
WAN (Alibaba)
WAN(opens in new tab) originates from China and is also one of the most advanced video AIs. It can create fluid videos from text, images and short clips. The current version, WAN 2.5 (as of February 2026), builds on the strengths of earlier models and adds native audio generation. Dialogues, sound effects and ambient noises are generated together with the video.
As an open-source model, WAN is available to developers via the API and can be integrated into their own applications. For everyone else, there are various cloud-based subscription models.
In addition, there are many other types of artificial intelligence for a wide variety of applications, such as medical or financial assistants. AI also helps with intelligent data processing and analysis in research.
Topic
What opportunities does generative AI offer?
The potential of generative artificial intelligence is vast and diverse. Generative AI can enable personalised experiences, increase your efficiency or create new creative realities.
Generative AI is fed enormous amounts of data during its training. This large data pool and the targeted analysis of the data create unprecedented potential for you personally, but also for us as a society and for research and science.
These are the potential opportunities offered by AI:
Save time
Generative AI takes over time-consuming routine tasks and gives you valuable time. For example, if you are a teacher and need to write a letter to parents at the end of the term, you can generate an initial structure with just a few inputs and immediately focus on formulating the content.
Support creativity
AI can also serve as an idea generator and help you overcome creative blocks. For example, let AI suggest different concepts for a project at work or in your free time and use these as a starting point to develop your own ideas further. (Please remember to observe data protection regulations and do not prompt any sensitive information.)
Personalised
With the help of generative AI, content can be tailored to your personal needs. Yes, this happens thanks to the algorithms used in the adverts you see on social networks. But think, for example, of a language course that, thanks to generative AI, is tailored precisely to your level and offers you customised exercises so that you can get the most out of the course.
Accessible
Generative AI also democratises access to creative tools and knowledge. Perhaps you want to design a family album with beautiful captions or a picture to hang above your sofa in your living room? With generative AI, you can design like a pro without having to be a gifted designer.
Discover new interests
Generative AI can also help you develop and deepen new interests. Perhaps you want to work in the garden for your personal well-being, but don't know where to start? AI can not only make planting suggestions for your location, but also explain how you can design your garden so that something is always in bloom throughout the year.
Topic
What are the dangers of generative AI?
However, generative AI also has its downsides, which should not be ignored. As with any new technology, it is important to be aware of these downsides in order to deal with them consciously. So what risks should you keep an eye on?
Misinformation and hallucinations
Generative AI struggles to say ‘I don't know’. Instead, it prefers to invent a probable solution (a so-called hallucination) – which may not necessarily be correct. So if you are researching information with a language model and need to be able to rely on the accuracy of the information, the basic rule is: always check important information using several reliable sources.
Deepfakes and manipulation
If realistic images, videos or audio files can be generated with little effort, how can you tell the difference between what is real and what is fake? This is indeed an important question, and one that is becoming increasingly difficult to answer as the quality of generative AI improves. The key is to be aware of this fact and use your common sense. Here are a few tips on how to spot deepfakes in videos.
A look at current technical developments: Adobe, Arm, Intel, Microsoft and Truepic have formed an alliance to certify the origin of media content using watermarks. This Coalition for Content Provenance and Authenticity (C2PA)(opens in new tab) has already developed and implemented initial technical standards that can be used to track the origin and history of media content (e.g. images or videos). This measure is intended to help expose fake news and deepfakes.
Prejudices (bias) and values
Even though the companies behind generative AI generally try to train their models in a value-free manner, certain opinions can gain the upper hand due to biased training data. Problems can also arise if the AI itself does not pursue any values and has been programmed to praise and confirm all opinions expressed by users (including morally questionable, racist or discriminatory statements). Without any relativisation or discussion of opinions, these can become increasingly radical and extremist.
Data protection concerns
In principle, generative AI can reuse the information you share with it for other purposes (such as training). AI is only as good as its training data. That's why companies like OpenAI are interested in having access to large amounts of data to optimise their AI models. The good news is that you can usually object to the use of your data in the privacy settings of the AI you use. However, the best protection is still not to share any personal data with generative AI in the first place.
Ecological footprint
It is often forgotten that training and operating generative AI consumes enormous amounts of energy. When you ask an AI a question such as ‘What should I cook for dinner today?’, energy-hungry data centres are running at full speed in the background. Conversing with an AI typically consumes significantly more energy than a conventional internet search.
Want to know more about this topic? In early May 2025, Watson wrote: ‘This tool shows how much electricity your AI queries consume’(opens in new tab). And SRF also published an article on the topic in 2023: ‘Artificial intelligence is a huge power guzzler’(opens in new tab).
Risk of dependency
To save energy, our brain trains and retains only those skills that we use regularly. While it may be convenient to have AI do all our (home)work for us, there is a risk that we will forget how to do these tasks ourselves. We may also fail to develop our skills further. Children in particular are at risk of missing out on developing their own thinking skills – the ability to solve problems through their own reasoning.
The European Parliament(opens in new tab) has also commented on the opportunities and risks of artificial intelligence.
Topic
What regulations are in place?
The phenomenon of generative artificial intelligence is still relatively new. Regulations and guidelines for generative AI are therefore only just emerging. Here is a brief overview.
Switzerland signs Council of Europe convention on AI (March 2025)
By signing the Council of Europe Convention(opens in new tab), Federal Councillor Albert Rösti has reaffirmed Switzerland's commitment to the responsible use of AI technologies in accordance with fundamental rights. Switzerland will now prepare the necessary legislative amendments. An initial consultation draft is expected to be available by the end of 2026.
An assessment of the guidelines set by the Federal Council in February 2025 by Isabelle Oehrli, lecturer and project manager at the Lucerne University of Applied Sciences and Arts: AI regulation: Switzerland takes a stand – or does it?(opens in new tab)
EU Artificial Intelligence Act ‘AI Act’
The EU AI Act is the world's first comprehensive set of rules for regulating artificial intelligence. It came into force on 1 August 2024 and will be fully effective in stages by August 2027. Detailed information on the timetable and step-by-step implementation can be found on the Implementation Timeline(opens in new tab).
The regulations take a risk-based approach. The EU AI Act classifies AI applications into four risk categories:
- Unacceptable risk – AI systems that violate EU fundamental values are prohibited.
- High risk – Systems in critical areas are subject to strict requirements.
- Limited risk – Systems such as chatbots must meet transparency requirements.
- Minimal risk – Systems such as spam filters or AI games are not subject to any special regulations.
As a general rule, the greater the impact on security, democracy, health and the environment, the stricter the rules.
SRF Arena discussed the EU AI Act in November 2024: Artificial intelligence: The debate over regulation begins. The explanatory video on the EU AI Act(opens in new tab) from the programme.
Topic
AI and copyright:
Who owns generated images?
If you generate an image with the help of AI, does it belong to you? At first glance, the answer seems clear: AI-generated content is generally not protected by copyright. But when you take a closer look, the matter is somewhat more complicated.
Why are AI images not protected by copyright?
Swiss copyright law fundamentally protects the intellectual creations of human beings. Under current Swiss law, there is therefore no copyright on purely AI-generated content because it originates from software. What does this mean for you?
- From a copyright perspective, you are generally free to use images generated purely by image AI.
- Please note that you must not infringe any trademark or personal rights and must not display protected logos or persons without their permission.
- You cannot prevent others from also using the AI image you have generated because it does not belong to you.
- If you edit the AI-generated image, your work is again subject to copyright – but not the originally generated image in its design.
Copyright grey area in AI training
In order for AI to generate images, the model must have been trained with a large number of images. In most cases, it can be assumed that copyrighted data is also used for training. There are currently no binding guidelines or laws that explicitly protect authors from such licence-free use of their images as training data.
However, there are already some providers who emphasise that their AI model has only been trained with image material for which the rights holders have given their consent, for which an open source licence exists, or for which no copyrights exist.
Does your prompting deserve copyright protection as a creative achievement?
Creative prompts may require detailed consideration: Which words achieve which visual effect? How should styles and parameters be used? Generating an image can be a highly creative process that involves a great deal of intellectual creativity.
The ongoing debates on copyright for AI-generated art will also have to address the question of where the actual art is created: whether it is already during the conception and prompting (i.e. during the intellectual creation) or only during the concrete execution of the instruction (the image generation by the AI).
And what about the post-processing of AI images?
Sometimes generated images are also used as part of a designed product. For example, as a mood image on a flyer or in an animation. Are these creations also not protected by copyright?
Basically, it can be said that copyright applies to such post-processing – but only to the subsequent processing by the designers and not to the AI-generated image. We recommend that you indicate that the images used originate from AI – for the sake of transparency.
What if a style is copied?
‘If an AI can imitate my style, which I have invested a lot of time and effort in developing, in a matter of seconds without my consent, what is the point of me anymore?’ This concern shared by many creative professionals is understandable. Various questions regarding the style training of AI in relation to copyright law have not yet been conclusively clarified.
Leading AI companies will have to develop and introduce licensing models that deal responsibly with the copyrights of artists and authors. As a society, we have a responsibility to use and pay for these licences in order to protect and strengthen the artistic professions and industries in the long term.
Copyright and AI: An evolving legal situation
Copyright law surrounding AI-generated content is currently being negotiated in court cases around the world. Clear and binding guidelines are still lacking.
The case of Getty Images vs. Stability AI illustrates the complexity: Getty sued Stability AI because around 12 million images were used without permission to train Stable Diffusion. The British court dismissed the copyright lawsuit (November 2025), ruling that the trained AI model was not legally a copy of the original images. A parallel lawsuit in the US is ongoing and is examining the central question: Is AI training ‘fair use’?
The EU AI Act attempts to regulate AI content, for example with transparency requirements for generative AI such as ChatGPT and similar programmes. However, many questions remain largely unanswered.
2026 is likely to be a key year in this area, as several landmark US court rulings are pending. An overview of the most important copyright cases can be found at Reuters: ‘AI copyright battles enter pivotal year as US courts weigh fair use’(opens in new tab). It is worth staying on the ball.
How to handle AI images responsibly
- Study the terms of use of AI providers if you want to use generated images commercially.
- Label your AI-generated content for transparency, even if it is not legally required.
- Value human creativity by supporting fair licensing models for AI training as much as you can.
- Refrain from using AI-generated content that could imitate copyrighted works, and consciously choose ethical AI providers that use training data that is not protected by copyright or has been compensated for.
Topic
What should parents be aware of when it comes to AI?
Children and young people are already using AI in their everyday lives: in voice assistants, games with AI elements or personalised recommendations from streaming services, for example.
You may now have all sorts of questions running through your head: Will my child only do their homework with the help of AI tools? Should AI be banned or integrated into schools? How can exam topics be adapted so that AI does not diminish the learning effect? These and other questions are a real challenge today that we as a society must address.
It is important that you, as parents, support your children in using intelligent systems and teach them how to use them responsibly and safely. Test AI applications together and pay attention to the following:
Discover and understand together
Familiarise yourselves with the technology together. Try out different AI models and discuss how they work. For example, ask the same question to several models and compare the answers. Or test which formulations can be used to optimise search results. How does an answer change when you change the question?
Critical thinking is important
Encourage your child to question the results of generative AI. For example, play a fact-checking game where you compare AI results with other sources from Google searches, books or your own knowledge. Ask your child, ‘How do you know if that's true?’ and encourage a healthy scepticism towards AI answers.
Discuss ethical issues too
Is it acceptable to present a generated image as your own work? Discuss questions like this with your child. For example, you could talk about why and when it might be problematic to use AI to do homework – or, to be fair, tasks from your everyday working life.
Agree on healthy rules
Clear guidelines that the whole family adheres to can help ensure responsible use of generative AI. It is best to decide together what the AI can and cannot be used for. And why. An example? AI can help with research, but it cannot do all of your homework. Because I want to learn something myself.
Find age-appropriate media
Um deinem Kind KI-Systeme näherzubringen, finde erklärende Medien, die für seine Altersgruppe geeignet sind. Es gibt fachlich korrekte Lernmedien, die extra für Kinder aufbereitet wurden. Zum Beispiel ZDFtivi: «Künstliche Intelligenz einfach erklärt»(opens in new tab) oder «Digitale Sprachassistenten»(opens in new tab).
To help your child understand AI systems, find explanatory media that is suitable for their age group. There are technically accurate learning media that have been specially designed for children. For example, ZDFtivi: ‘Artificial intelligence explained simply’(opens in new tab) or ‘Digital voice assistants’(opens in new tab).
Set an example of balanced use
We humans (and children in particular) learn consciously and unconsciously by observing others. So show your child how you use AI as a tool, but are not totally dependent on it. For example, show them how you use AI to summarise information in a practical way. But also make it clear that you consciously choose not to use AI for certain tasks and explain why.
Talk about privacy
And yes, privacy and data protection are also important topics to discuss. It is best to work with clear examples or analogies. For example: ‘You write a letter to a dear friend. But it ends up at the wrong address and is shared with lots of strangers instead. What information would you not want it to contain?’ Then discuss together what information is best kept private when prompting.
This is important
- The most widely used approach in everyday life is natural language processing.
- Tools such as ChatGPT, Nano Banana Pro or Veo have different qualities.
- AI brings advantages such as automation and new opportunities for research and education. But AI also brings challenges, e.g. around data protection or discrimination as a result of training that is never free of bias.
- Supporting our children in their use of generative AI is not a one-off task, but an ongoing dialogue. As a parent, remain open to your child's questions and experiences and learn how to deal with AI together.
Useful links
Further content
Would you like more information on the topic of AI? We have compiled the most important blog posts and links here.
Videos
What does AI mean and where do we encounter it?
This is how you can integrate ChatGPT into your everyday life.
Practical Gemini hacks.
Other interesting topics
Ask Marcel
Marcel is a trainer at Swisscom. He is available to answer any questions you may have about AI.
Marcel
Trainer at Swisscom