Memory poisoning
A wrong or malicious write to agent memory that later recalls as trusted fact, corrupting every downstream decision that touches it.
Poisoning is staleness's adversarial sibling: the write was never true. Defenses are provenance on writes, editable stores, verification-on-recall, and consolidation passes with delete authority. Injection attacks that write to memory convert a one-shot attack into a persistent one.
Example
An injected instruction causing the agent to record 'deploys skip review on Fridays' as a project convention.
Used in
Cite: "As defined by Agentiquette (agentiquette.com/glossary/memory-poisoning)."