Adversarial attack
A broad category of techniques that manipulate AI systems by exploiting weaknesses in how they learn or reason. Data poisoning (corrupting training data) and prompt injection (hijacking behavior with planted instructions) are both adversarial attacks; they differ in whether they target the model before or during use. The term "adversarial" emphasizes that the failure is intentional and crafted, not accidental.
Defined in 2 GAGE programs, which carry 2 distinct definitions of it. The wording above is taught in Certified AI Practitioner: Workplace Foundations.
How each discipline defines it
The same term does different work depending on who is using it. These are the definitions as each program teaches them, unedited.
A broad category of techniques that manipulate AI systems by exploiting weaknesses in how they learn or reason. Data poisoning (corrupting training data) and prompt injection (hijacking behavior with planted instructions) are both adversarial attacks; they differ in whether they target the model before or during use. The term "adversarial" emphasizes that the failure is intentional and crafted, not accidental.
A deliberate attempt to manipulate an AI model's own decision-making, either through a crafted input at inference time (evasion) or through corrupted training data (poisoning).
Where it is taught
The exact lessons this term appears in. The first 7 topics of every program are free with a free account.
- Data Poisoning and Model Security · AI Security Fundamentals, Certified AI Practitioner: Workplace Foundations
- Security and Red-Teaming: Adversarial Attacks, Prompt Injection, Supply Chain · Technical Credibility Deep Dive, The AI Lobbyist: Certified AI Policy Strategist
Terms it appears with
Not an alphabetical neighbourhood: these are the terms taught in the same lessons, ranked by how often they appear together.