Why AI concerns you

What this is about

AI already helps decide what you see when you search, who reads your job application and who answers your complaint. Mostly without you ever having asked for it.

This guide keeps asking two questions:

• Privacy: Who sees what you type in – and what happens to it afterwards?
• Agency: Who makes the final call – you, a person, or a system you can't see?

The examples below show where that already matters today. Each one leads to a topic covered in more depth later.

When software screens out your job application

Many companies have CVs pre-sorted before a person reads them. According to the report "Artificial Intelligence in Human Resource Management" by the AI service centre of RTR, Austria's communications regulator (April 2025), around two thirds of HR departments in companies with more than 50 employees already use AI, for example to pre-select applications (Der Standard, in German). If you're filtered out there, you usually won't even learn that software made the call.

Agency: The GDPR already protects you. Under Article 22, a decision with significant effects may not be taken by automated means alone; where an exception applies, you're entitled to human intervention and to contest the decision (GDPR, Art. 22). A person who merely rubber-stamps the result is not real oversight, according to the European Data Protection Supervisor (EDPS, TechDispatch 2/2025).

Privacy: AI video interviews analyse your face, voice and choice of words – data you would never hand over in a normal conversation.

The stricter rules of the EU AI Act for recruitment software only apply from 2 December 2027 (Regulation (EU) 2026/1744). According to the Chamber of Labour, that doesn't leave a legal vacuum: the GDPR and Austria's Labour Constitution Act already apply (AK Wien, in German).

When customer service is a chatbot

Since 2 August 2026, a chatbot in the EU must tell you that it is an AI – unless that's obvious anyway (AI Act, Art. 50). The Chamber of Labour uses exactly this example: with a customer-service chatbot, it must be clear that no person is answering (AK Wien). If in doubt: ask it directly.

Agency: There is no general right yet to speak to a person. Exceptions exist for consumer credit (Consumer Credit Directive (EU) 2023/2225) and for financial services sold at a distance (Directive (EU) 2023/2673). That may change soon: the European Commission plans to present a draft Digital Fairness Act on 11 November 2026 (CMS: "Inside the EU's planned Digital Fairness Act"), and a right to reach a person when making a complaint is seen as a likely part of it (A&L Goodbody: "Digital Fairness Act: A necessary safeguard or unnecessary duplication?").

Privacy: What you type into the chat is usually stored. Account numbers, health information or details about other people belong there only when truly necessary.

When the search engine answers for you

Google puts an AI summary above the results, and it can't be switched off. A test by the company Oumi, commissioned by the New York Times and covering 1,609 search queries in early 2026, found that 91% of the summaries contained the correct answer – but only 39% were both correct and backed in every statement by the linked sources (Oumi: "Oumi's Study Finds 50% of AI Overviews Untrustworthy"; New York Times: "How Accurate Are Google's A.I. Overviews?", 7 April 2026). Google disputes the method (Newsweek).

Agency: The summary decides for you which sources count. The link next to it is no proof that the answer is right.

Privacy: AI answers can also be wrong about people. That's why the Vienna-based privacy organisation noyb filed complaints against OpenAI in 2024 and 2025: ChatGPT had made false claims about people, up to an invented murder of the person's own children, and OpenAI stated it could not correct false information (noyb: "ChatGPT provides false information about people, and OpenAI can't correct it"; TechCrunch). Under Article 16 GDPR, you have a right to have false data about you corrected.

Covered in more depth later: Cross-checking – why a disagreement between two answers is the most important signal, and how to settle it against sources.

When data gets pasted into a chatbot

A customer letter, a sick note, a salary list – quickly pasted into a chatbot to draft a reply. With that, the data leaves the building, going to a provider whose terms nobody at work ever negotiated.

Privacy: It's not just you who's affected, but people who never agreed to it: customers, colleagues, members.

Agency: Whether and how AI is used at work is not just an IT question. In Austria, the works council has a right to know whether and where AI is used (section 91 ArbVG), and systems that process employees' data need its consent (section 96a ArbVG) (Labour Constitution Act in RIS, German; AK Wien: information and co-determination on AI, PDF in German). AI-based emotion recognition at the workplace is banned in principle under the AI Act (AK Wien).

The union GPA additionally calls for a right to a "human interface" in every company that explains AI results, and for an easy-to-understand label whenever employees' data is passed to AI systems (GPA Federal Forum 2025: motion on AI, PDF in German).

Covered in more depth later: Masking data – what anonymisation achieves, what it doesn't, and what works councils should ask before such a tool is introduced. Plus a blueprint of what a protected AI setup looks like in principle.

When the AI agrees with you

AI chatbots tend to agree with you. In April 2025, OpenAI pulled a ChatGPT update after four days because it told users what they wanted to hear – up to endorsing harmful ideas (OpenAI: "Sycophancy in GPT-4o"; Georgetown Law: "Tech Brief: AI Sycophancy & OpenAI").

Agency: An answer that sounds like your own opinion feels like confirmation. Often it's just an echo, and sounding confident doesn't make it right.

Covered in more depth later: Thinking traps – why fixed routines protect better than attention. And testing models yourself instead of trusting other people's rankings.

Help and where to turn

For most of the situations above, there's an organisation in Austria that can help.

What comes next

The sections will appear here on gstir.net one by one; the names below will then become links. Each carries its own "as of" date, because rules and tools change quickly.

This page gives general information and is not legal advice. The graphics are my own; the data behind them comes from the sources named in each caption.

How this page was made

This guide recommends transparency about AI use – so it practises it.

The graphics are AI-made depictions of publicly available data; the masking example is invented. AI makes mistakes, even after checking. If you find one: dominique@gstir.net.