Rowhammer attack can backdoor AI models with one devastating bit flip

Rowhammer attack can backdoor AI models with one devastating bit flip

Rowhammer exploits reveal that flipping a single memory bit can permanently poison AI models, bypassing standard defenses. This vulnerability threatens critical infrastructure, allowing attackers to manipulate autonomous vehicles or facial recognition systems with undetectable precision. The findings underscore the urgent need to secure hardware memory in open data ecosystems, as compromised physical layers directly invalidate the integrity of publicly shared datasets and deployed machine learning models.

Related news

• UCR pioneers way to remove private data from AI models
• “Las obras creadas por IA no pueden registrarse como derechos de autor”: INDAUTOR
• Copyright challenges in AI model training under Thailand’s legal framework
• Pundi AI Brings AI Data Economy to BNB Chain with THENA Finance Listing
• How Ambani plans to milk the Kamdhenu cow