Is AI human-ready?
summary
State-of-the-art classifiers flip from “pig” to “airliner” with 99% confidence after a small, carefully chosen noise pattern, and MIT students print a turtle that humans read as turtle but models label as rifle from most angles. Face recognition systems change identity when a person wears specially designed glasses. These are concrete adversarial perturbations, reproducible attacks that expose security risks and safety failures in real deployments and motivate robust defenses and verification tools that actually work under realistic threat models.
AI-written summary · about
chapters (5)
related talks
44mComputers are Stupid: Protecting "AI" from Itself
46mA Short Introduction to Hacking Artificial Intelligence
35mHacking AI Systems: How to (Still) Trick Artificial Intelligence
19mAI: The current state of play
01h 05mNavigating the Security and Privacy Landscape of Modern AI
20mMachine Learning Attacks – A New Era of Security Epidemic
42mIs your AI plotting to kill you? Why AI and Machine Learning is dangerous
50mSecurity as Part of Responsible AI: At Home or At Odds?