Diffusion models are vulnerable to supply-chain PRNG hijacking that forces pixel-perfect attacker-chosen outputs, and QRNG eliminates the attack.
Guardt2i: Defending text-to-image models from adversarial prompts,
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CR 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
DiffusionHijack: Supply-Chain PRNG Backdoor Attack on Diffusion Models and Quantum Random Number Defense
Diffusion models are vulnerable to supply-chain PRNG hijacking that forces pixel-perfect attacker-chosen outputs, and QRNG eliminates the attack.