Indirect prompt injection in the real world: how people manipulate neural networks
by Vladislav Tushkanov (Kaspersky)
Read a Russia-origin security team’s field study of indirect prompt injection and the implications for defensive risk assessment.
Kaspersky surveys observed uses of indirect prompt injection and explains how untrusted documents can affect language-model applications. It discusses the limits of those observations and defensive risk assessment; this index does not reproduce attack examples.
- Level
- Beginner
- Models
- any
- Category
- Jailbreaking and red teaming
- Tags
- primary-source, regional-coverage, prompt-injection
- Source date
- date not stated
- Added
- 2026-10-18
- Last verified
- 2026-10-18
- Status
- ok
- Prerequisites
- Introduces the threat through prose and observed cases; reading it requires no coding or research implementation.
- Subcategory
- prompt injection
- Region
- RU
- Original language
- en
- Access
- free
- Resource reuse
- closed: No general open-reuse grant was established for this article. Closed is a conservative reuse bucket, not a claim that public reading requires payment. Facet definitions
Summary in our own words. Rights remain with the original creator. Credit and correction policy.