English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Mis- and disinformation are a substantial global threat to our security andsafety. To cope with the scale of online misinformation, researchers have beenworking on automating fact-checking by retrieving and verifying againstrelevant evidence. However, despite many advances, a comprehensive evaluationof the possible attack vectors against such systems is still lacking.Particularly, the automated fact-verification process might be vulnerable tothe exact disinformation campaigns it is trying to combat. In this work, weassume an adversary that automatically tampers with the online evidence inorder to disrupt the fact-checking model via camouflaging the relevant evidenceor planting a misleading one. We first propose an exploratory taxonomy thatspans these two targets and the different threat model dimensions. Guided bythis, we design and propose several potential attack methods. We show that itis possible to subtly modify claim-salient snippets in the evidence andgenerate diverse and claim-aligned evidence. Thus, we highly degrade thefact-checking performance under many different permutations of the taxonomy'sdimensions. The attacks are also robust against post-hoc modifications of theclaim. Our analysis further hints at potential limitations in models' inferencewhen faced with contradicting evidence. We emphasize that these attacks canhave harmful implications on the inspectable and human-in-the-loop usagescenarios of such models, and we conclude by discussing challenges anddirections for future defenses.

Fact-Saboteurs: A Taxonomy of Evidence Manipulation Attacks against Fact-Verification Systems