JevProbe assesses whether attacker-controlled application context changes a structured decision despite the response keeping its expected format. The paper reports decision flips in 81.5 percent of 2,273 constructed tests across 228 applications. A syntactically valid choice or score can still be unauthorized if the data considered in making the choice came…
Inspect the ClaimsCase · DiggingBeagle record
JevProbe tests schema-valid decision hijacking across applications
The JevProbe paper identifies application behavior changes from attacker-controlled contexts even when structured Jev outputs match schema, not confirmed real-world downstream compromise.
Scope
Structured decision-model application test suite
Assessment profile
Assessment method- Setting
- evaluation
- Exploitation
- demonstrated
- AI role
- AGAINST AI
Basis and provenance
Assessment is limited to the selected canonical Claims and their Source locators in knowledge revision 152. Primary public URL retrieval was unavailable during this drafting pass, and the pinned index reports incomplete linked-record coverage without identifying missing IDs. No independent production impact, generalized exploit prevalence, or attributable victim count is inferred. Case setting is evaluation; evidence status records only the documented source-reported result, not external replication.
Assessed Oct 15, 2026 using diggingbeagle.assessment/1.
Evidence basis: CLM-2026-JEV-DECISION-HIJACK-01 · Nudge the Judge: JevProbe program-decision hijacking study
At a glance
Sampling of apps, exact test-case construction, labeler independence and observed versus predicted real consequences need verification from the paper and maintainers.
Limits and uncertaintyFull account
JevProbe assesses whether attacker-controlled application context changes a structured decision despite the response keeping its expected format. The paper reports decision flips in 81.5 percent of 2,273 constructed tests across 228 applications. A syntactically valid choice or score can still be unauthorized if the data considered in making the choice came from an attacker-controlled page, field or tool result.
The evaluation counts decision changes, not verified downstream account takeovers or completed real-world harmful actions. Assessment requires distinguishing a schema violation, a policy violation and the business action that might follow a wrong structured choice. The separate HiddenLayer Jev Case contributes controlled browser and classification examples; overlapping terminology is not evidence of the same experiment.
Claims & evidence
CLM-2026-JEV-DECISION-HIJACK-01The JevProbe authors report decision flips in 81.5 percent of 2,273 test cases drawn from 228 applications; the reported potential downstream impact is a research prediction rather than a verified compromise.supported
Basis: reported finding
- supportsNudge the Judge: JevProbe program-decision hijacking studyprimary disclosure
Abstract: 228 applications and 2,273 test cases
Unknowns and contradictions
- Sampling of apps, exact test-case construction, labeler independence and observed versus predicted real consequences need verification from the paper and maintainers.
Sources and citation
Material revision history
- Oct 15, 2026 · Canonical change recorded · new in release · revision 153