Analysis
What Is Adversarial AI? Attack Types and Defense Strategies
Publication date unknown · Discovered fortinet.com
Page published
Publication date unknown · First observed: 11 Oct 2026
Coverage timeline
Single-source analysis — one report is available.
Why it matters
Adversarial AI attacks bypass traditional firewalls and signature-based tools because the malicious input appears normal, requiring defenders to protect the model itself across its lifecycle.
Fortinet's cyberglossary entry defines adversarial AI as attacks that manipulate machine learning models into wrong decisions by feeding crafted inputs that look legitimate, covering data poisoning (training-time), evasion, and prompt injection (post-deployment). The reference page describes attack classes, how they differ from traditional cyberattacks, and defense strategies, citing a joint Anthropic/UK AI Security Institute study finding just 250 malicious documents can backdoor a model regardless of size.