The AI Taxonomy.
Terms  /  Memory & State  /  State Loss
03 · Memory & State

State Loss

Failure to maintain awareness of prior actions.

State Loss is simpler and more immediate than Context Anxiety — it happens within a single session, not across sessions. The model literally loses track of what it just did: it forgets that it already created a file, re-implements a function it wrote ten turns ago, or contradicts a decision it made five minutes earlier. State Loss is one of the primary drivers of Execution Hallucination, because a model that can't track its own actions can't accurately report on them.

Example
An agent is tasked with refactoring a module. At turn 12, it renames a function from `processData` to `transformInput`. At turn 18, it writes new code that calls `processData` — the old name — because it's lost track of the rename. The resulting code has a broken reference. When the user points it out, the agent "remembers" and fixes it, but the damage is done: the user now has to verify every change the agent makes because they can't trust it to remember its own work.

More in Memory & State

Context Anxiety

The dread of re-explaining everything every session because the system has no persistent memory.

Ghost Knowledge

Facts that exist in the system but can't be traced to an author or source.

Context Rot

Deterioration of an agent's understanding over time due to limited memory and compaction.

Context Continuity

Seamless persistence of knowledge across sessions — the user never re-explains. The cure for Context Anxiety.

Memory Fidelity

The accuracy and completeness of what a system retains over time. The opposite of Context Rot.

Provenance Clarity

Every fact in the system is traceable to its author, source, and reasoning. The cure for Ghost Knowledge.

Recall Precision

The system retrieves exactly the right context at the right time — no noise, no gaps.

State Awareness

The system's ability to maintain and use awareness of prior actions within a session. The opposite of State Loss.