Skip to content
SwissAnon
Contact
Menu
Back to blog
2026 keynote talk

Keynote at the Austrian Statistics Days 2026 in Vienna: disclosure scenarios, attacker cost and agentic AI

On 9 September 2026, Matthias Templ (FHNW, SwissAnon) gave the keynote at the Austrian Statistics Days 2026 (Österreichische Statistiktage) of the Austrian Statistical Society, held at Statistics Austria in Vienna. The talk, Disclosure scenarios, attacker cost, and agentic AI, opened Session 3, “Innovative Methods in Official Statistics”.

The room was full, and the talk was followed by a lively discussion.

The argument of the talk ran in three steps. First, disclosure risk is not a property of a dataset but of a scenario: who the attacker is, what they already know and which sources they can link. Second, every scenario carries an implicit price. Much of what is treated as anonymous in practice rests on the assumption that re-identification would take more effort than anyone would spend. Third, that price has collapsed. In our study with simulated location traces of consenting volunteers, a team of AI agents named 18 of the 25 identifiable people from public sources alone, at about 2.24 dollars and 17 minutes per person.

The talk also covered the measurement side: how classical anonymisation and synthetic data move a release on the risk-utility map, why no single method is both safe and useful, and why the choice of where to stand on that map is a governance decision, not a statistical one.

Many thanks to Alexander Kowarik (Statistics Austria) for organising the session and for the invitation to give the keynote.