Embedding AI-confusing elements into original images to block generation

Seoul — Lavarwave, a company developing digital crime response technology, announced on the 23rd that it has jointly developed a technology to prevent AI-powered deepfakes with the KAIST Cybersecurity Research Center.

Lavarwave and the KAIST Cybersecurity Research Center developed a method that protects original images so deepfake results cannot be generated, using adversarial attack techniques against generative AI.

The deepfake prevention technology adds imperceptible micro-level noise to original images. While ordinary users cannot detect any difference, the noise causes critical confusion for AI models, preventing them from generating deepfake content.

Existing deepfake detection technologies or digital watermarking mainly serve as passive defenses that detect or track already-generated content. In contrast, this new approach is described as a proactive method to prevent deepfake harm before it occurs.

Lavarwave stated that the deepfake prevention technology is highly versatile and can be applied across various platforms and devices. The company expects demand from smartphone manufacturers, social media platforms, cloud service providers, and the public sector.