evidoria

← Back to browse

Good practice Imported

AMS Berufsinfomat — Austria's Public Career-Information Chatbot and Its Bias Scandal

Austria · Vienna · See the Austria profile · See the Vienna profile

Evidence: Descriptive / self-reported Top 70% 40/100 · Ask Evidence Copilot about this practice

Austria's public employment service runs a free generative-AI chatbot answering career, salary and training questions, used in schools and job centres. Days after its January 2024 launch it produced gender-stereotyped and discriminatory advice, forcing rapid retrofits.

~140,000
Questions asked (first 2.5 weeks) (January 2024)
122,150
Answers generated in 2025 (2025)
275,160 EUR
Development cost to 31 Dec 2023
~5,000
Questions flagged for gender bias (of first 140,000)
1,857,200
Karrierekompass visitors (2025, for scale) (2025)
AMS Berufsinfomat — Austria's Public Career-Information Chatbot and Its Bias Scandal

Details

Maturity
Established
Promoter
Arbeitsmarktservice Österreich (AMS)
Period
2023–ongoing
Region (NUTS)
AT13
Keywords
career guidance, public employment services, generative AI, school-to-work transition, chatbot

Context

The Berufsinfomat is a free generative-AI chatbot launched 3 January 2024 by Austria's public employment service AMS, retrieving from AMS's own career-information content and passing it to a commercial LLM via the OpenAI API.

Activities

AMS built bias-mitigation into its process after launch: it flagged around 5,000 of the first 140,000 questions as prone to gender bias, built a 1,000-question gender test set, and ran regression tests, shipping corrections on 5, 10 and 19 January 2024.

Results

Roughly 140,000 questions were asked in the first 2.5 weeks; AMS reports 122,150 answers generated during 2025; development costs to 31 December 2023 came to EUR 275,160.

Conclusions

Within 48 hours of launch the tool became a national controversy over gender-stereotyped and discriminatory answers; there is no impact evaluation, controlled study, or published user-satisfaction data, and Austria already carried a related AMAS profiling-algorithm bias precedent.

Implementation
Indicative cost
Low (< €50k)
Time to results
Medium (1–3 years)
Staffing & skills
GoodGuys GmbH, Bundesrechenzentrum GmbH, Braintrust Marketing Service GmbH, AMS

Conditions for success

  • Regression testing on bias after each correction
  • Transparent procurement disclosure

Common failure modes

  • Gender-stereotyped and discriminatory answers documented within 48 hours of launch
  • Direct-award procurement without competing tenders
  • Source code obfuscated after criticism

Commonly funded by

National / regional programmes ESF+ — European Social Fund Plus

Indicative funding routes for practices of this type — always check each programme's current calls and eligibility rules.

Do you run this practice? Claim it — verified implementers get a public contact pathway and can propose corrections.

Data sources

Where this practice's information was retrieved from, and when.

Attachments

Similar practices you may find useful