Evidence · DiggingBeagle record
JevAdvBench: Black-Box Attacks on Typed Decision Models
Adversarial benchmark of typed probabilistic, ordinal and categorical AI decisions. Attacker-controlled factual ordering and persuasion can change valid structured outputs in tests, not prove live service compromise.
- Source role
- primary disclosure
Evidence record
Adversarial benchmark of typed probabilistic, ordinal and categorical AI decisions. Attacker-controlled factual ordering and persuasion can change valid structured outputs in tests, not prove live service compromise.
Claim-level citations (2)
- supportsTyped AI decision outputs remained vulnerable to persuasive adversarial input: JevAdvBench evaluates black-box perturbations against typed models that return probabilities, choices or ordered levels, with downstream services routing automatically above decision thresholds.
Abstract; threat model and downstream service
- supportsTyped AI decision outputs remained vulnerable to persuasive adversarial input: The paper’s black-box suite tests argument-driven and evidence-order inputs, not only imperative instructions, and tracks whether answers remain valid in the required typed format.
Section 3 benchmark attack suite; Section 4 attack results