Noise Poisoning (2024-) is a long-term project where imperceptible adversarial noise is injected into training images used by commercial AI systems. The corrupted datasets disrupt machine learning models, producing cascading errors that destabilize supposedly “neutral” algorithmic perception.
In this work, Varga reframes digital sabotage as both an ethical and aesthetic practice: a strategy of reclaiming agency within systems that otherwise absorb and reproduce everything they see. Each poisoned image is a quiet insurgency—a minimal gesture that spreads unpredictably through vast networks.
Rather than portraying resistance from the outside, Noise Poisoning operates within the system’s blind spots. What begins as a technical interference unfolds into a meditation on fragility, agency, and the beauty of machine failure.