Austria's Union Movement Draws a Red Line on AI That Watches Workers
Published on 11/06/2026 at 15:04 | Editorial boerse-global.de
Austria's largest labour federation is pushing for binding legal limits on artificial intelligence in the workplace, arguing that what is technically possible should not automatically be treated as lawful. The Österreichische Gewerkschaftsbund (ÖGB), which says it represents roughly 1.2 million employees across the country, is reacting to a newly introduced analytics feature from US AI developer Anthropic that lets companies scrutinise how staff work and which software they use.
At the centre of the dispute is "Smart Reports", a tool Anthropic offers to large corporate clients. Administrators can generate up to ten free reports a month, selecting specific teams and time windows of up to 28 days. The Claude language model then examines prompt logs, summarises task areas, sets token costs against the work done and flags apparent inefficiencies — among them meetings judged to have no direct work value and personal activities.
Names Can Be Unmasked, With a Log Entry
Anthropic stresses in its own help pages that the feature is not designed for performance reviews or employment-law personnel decisions. It is switched off by default, and usernames are hidden. Even so, authorised managers can manually reveal real names attached to session summaries, and the system records that action.
According to the provider, conversations touching sensitive areas such as health, union activity, religion and politics are excluded from analysis, as is content marked private. Enterprise customers account for around 80 percent of Anthropic's revenue.
Sebastian Klocker, the ÖGB's digitalisation expert, said technical feasibility must never be equated with legal permissibility. AI, he argued, must not be allowed to act as a digital minder in the workplace. The union body wants works councils involved early — before relevant software is even purchased — and full transparency about why company data is being collected.
Existing Rules Already Apply, Union Says
In making its case, the ÖGB points to rules that are already on the books. Under the General Data Protection Regulation (GDPR), employees are protected against purely automated decisions that have significant effects on them.
The EU AI Act adds further guardrails: systems used for hiring, promotion, dismissal or for assessing performance and behaviour count as high-risk AI. Using AI to recognise emotions at work is banned outright, permitted only in narrowly defined exceptions for safety purposes or medical care.
Company-level labour law also comes into play. Where a works council exists, any monitoring measure capable of affecting human dignity — much like video surveillance or GPS tracking — requires a works agreement (Betriebsvereinbarung) before it can go ahead.
From Sandwich Chains to Investment Banks
Real-world deployments are fuelling the debate. British retailer The Co-op, according to media reports, uses an OpenAI model to analyse staff conversations with customers word for word against more than 50 parameters. The company describes the system as an aid; unions criticise a lack of accountability.
Burger King has been reported to use AI-linked earpieces in its restaurants to assess how service-minded staff are and to issue automated instructions. Meta reportedly captures keystrokes and screen content to train its own models.
Rollouts are not always smooth. Industry association Bitkom found that around a third of German companies were caught off guard by rising costs for AI applications. An AWS study shows that only 15 percent of firms in Germany now deploy AI so extensively that it generates new products or processes — down from 21 percent the previous year.
Some large corporations are also pulling back. Investment bank Goldman Sachs largely blocked access to Claude Code and GitHub Copilot, restricting use to selected software engineers.
