As cited
Copy frozen at (site build).
ai security
Hackers could poison AI models by tampering with core infrastructure
Adversaries could compromise artificial intelligence (AI) models by tampering with hardware components in the infrastructure that trains or deploys them, potentially altering model weights and behavior. The attack surface extends across the entire supply chain of computational systems that support AI systems rather than targeting only software layers.
Why it matters: Organizations deploying or training AI models need to evaluate hardware security, supply chain integrity, and infrastructure monitoring to detect unauthorized changes that could degrade or poison model outputs.
- Source published
- First seen by Cybersecurity Tracker