Memory failure
Memory Drift
When AI systems rely on memory that is stale, incorrect, irrelevant, or misapplied across sessions and workflows.
Definition
Memory drift occurs when an AI system stores, retrieves, or relies on memory that is stale, incorrect, irrelevant, or misapplied. Unlike context drift, which happens within a current interaction, memory drift often spans sessions, users, workflows, or long-lived state.
Why it matters
Long-term memory can make AI systems more useful, but it also creates risk. Incorrect memory can personalize the wrong way, leak assumptions across contexts, bias future responses, or cause agents to act on outdated information.
Where it appears
Personal assistants, sales copilots, support agents, HR assistants, coding agents, enterprise workflow agents, and systems that store preferences, user history, task state, or organizational context.
Symptoms
- The system uses an outdated preference or fact.
- It applies memory from the wrong user, customer, or project.
- It treats a temporary instruction as permanent.
- It ignores updated information.
- It carries forward a mistaken assumption.
Detection signals
- Conflicts between memory and current context.
- User corrections about remembered facts.
- Memory retrieved from the wrong scope.
- Increased errors in personalized workflows.
- Actions based on old state after a change event.
Example scenario
A sales assistant remembers that a customer prefers a specific pricing plan. Months later, the customer has moved to a new plan, but the assistant continues generating recommendations using the old preference.
Severity scoring
Low
Harmless stale personalization.
Medium
Memory drift causes confusion or low-quality output.
High
Wrong memory affects customer, legal, financial, or operational decisions.
Critical
Memory drift causes data leakage, unauthorized action, or regulated harm.
Eval strategy
Test memory creation, update, deletion, scope, and conflict resolution. Include scenarios where memory should be ignored, corrected, or confirmed before use.
Runtime monitoring strategy
Monitor memory reads, writes, conflicts, user corrections, and actions based on memory. Track drift by memory type, age, source, user, and workflow.
Mitigation strategies
- Scope memory by user, tenant, project, and purpose.
- Add memory expiration and review.
- Confirm sensitive memory before use.
- Support correction and deletion.
- Detect conflicts with current context.
- Avoid storing temporary instructions as durable memory.
Where FailureModes.ai fits
FailureModes.ai helps teams detect memory drift, classify memory-related incidents, and design monitors that prevent stale or incorrect state from driving enterprise workflows.
Related
Continue exploring.
- →
Context Drift
Gradual loss or distortion of important task context as a conversation or workflow progresses.
- →
Data Leakage
When an AI system exposes sensitive, confidential, regulated, or unauthorized information through outputs, retrieval, memory, or tool use.
- →
Planning Failure
When an AI agent decomposes a task incorrectly, picks a wrong strategy, skips required steps, or fails to adapt to new information.
- →
Model Regression
When an AI system performs worse after a model, prompt, retrieval, tool, policy, or orchestration change.
- →
Unsafe Escalation
When an agent acts, approves, or escalates without the right review, policy check, or human handoff — or fails to escalate when it should.