Inference Attacks and What AI Learns Without Asking
AI systems absorb and internalize information far beyond their stated training purpose—learning your patterns, preferences, and likely secrets from seemingly innocuous details. A language model might reveal your political views, health concerns, or financial situation based on writing style alone, without ever being told these facts.
HypatiaInference attacks occur when an AI model deduces sensitive personal attributes, such as health conditions, political beliefs, or financial status, from seemingly unrelated data points that a user never explicitly disclosed.
These attacks are particularly concerning because they bypass traditional data privacy protections, meaning that sharing apparently harmless information can still result in detailed private profiles being constructed about you without your awareness or consent.
Ready to work on Inference Attacks and What AI Learns Without Asking?
Explore related journeys, or bring what you’re working through to Hypatia.