Skip to main content
Core concept · Emerging

​
Definition

Context poisoning is a form of adversarial attack on AI systems in which malicious content is injected into the retrieval context — through prompt injection in retrieved documents, manipulated knowledge base entries, or contaminated external sources — to cause the AI system to generate false, misleading, or harmful outputs. Context poisoning is primarily a security concern rather than an optimization concern, but it has practical implications for brands. A brand’s Wikidata entry, Wikipedia article, or key third-party descriptions are potential vectors for context poisoning if they can be edited by malicious actors. Monitoring these sources for unauthorized or inaccurate edits — and ensuring that the brand’s authoritative sources are well-maintained — reduces the risk of poisoned context corrupting AI representations of the brand.

Knowledge graph poisoning

Entity injection

Brand grounding

Hallucination mitigation

Synthetic brand signal

​
Relevant Plate Lunch Collective Services

Entity SEO Context Map