Subtle adversarial image perturbations can steer commercial AI assistants toward designated wrong answers, and the resulting answer patterns can statistically flag students who blindly copy those assistants.
Under the conditionally independent Bernoulli model, if the true null probabilities satisfy qfp i,j≤qfp i,j, calibrating the upper-tail threshold underqfp j is conservative
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.AI 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Fighting Fire with Fire: On the Feasibility of Protecting Exercises Against AI Cheating
Subtle adversarial image perturbations can steer commercial AI assistants toward designated wrong answers, and the resulting answer patterns can statistically flag students who blindly copy those assistants.