Hackers Could Poison AI Models Through Core Hardware
Researchers warn that attackers could tamper with the basic infrastructure that AI systems run on. Doing so could quietly corrupt a model so it behaves wrongly. Companies and users who rely on these systems could be affected, so the risk is worth watching for anyone building or using AI tools.
Quick glossarymodel poisoning, core infrastructure
- model poisoning
- Secretly corrupting the data or files an AI model learns from so that it later behaves in a harmful or wrong way.
- core infrastructure
- The basic hardware and software that a system depends on to work, like the foundation under a building.
Burning Potato score: 67%