Safety, Ethics & Governance · advanced · concept 291 of 311 · Updated Nov 4, 2026
Adversarial Machine Learning
Adversarial machine learning studies how attackers fool, steal from or corrupt models, and how to defend them. Small crafted changes can make a vision model misread a stop sign, and crafted text can push a language model past its rules. Attacks transfer between models more often than people expect, so this belongs in security planning rather than only research.
Top AI contributors on this
- Geoffrey HintonUniversity of Toronto (emeritus)
- Yoshua BengioLawZero / Université de Montréal
- Ilya SutskeverSafe Superintelligence Inc. (SSI)
- Dario AmodeiAnthropic
- Stuart RussellUC Berkeley
- John SchulmanThinking Machines Lab
Key terms
Evasion attacksModel extractionRobustnessAdversarial examplesThreat model
Learn these first
Where you meet it in the real world
Security testing of vision systems, fraud models and LLM products
Guides and articles
Adversarial Attacks on LLMs ↗
Lilian Weng
Explaining and Harnessing Adversarial Examples ↗
Ian Goodfellow and others
This unlocks