Mechanisms
Memory Processes · Acquisition and Transformation
Reconsolidation
Example
You have not thought about your first apartment in years. A friend mentions the street name and the place comes back: the layout, the noise from the corner, the walk to the bus. Over the next hour she fills in things you never knew at the time, like why the upstairs tenants left so suddenly. Months later, when the apartment comes to mind again, her account is simply part of what you remember about living there, not a separate thing you were told about it afterward.
Reconsolidation in Humans
Definition
Retrieving a consolidated memory returns it to an unstable state, from which it must be rebuilt to persist. During that window the trace can be altered or lost, so remembering is not a read-only operation (Nader et al., 2000).
Neural Substrate
The causal evidence is rodent. Nader et al. (2000) infused the protein synthesis inhibitor anisomycin into the lateral and basal nuclei of the amygdala after reactivating an auditory fear memory in rats, and found amnesia on later tests; the same infusion without reactivation left the memory intact, and the effect held whether reactivation came 1 or 14 days after conditioning. The window is narrow and the deficit is delayed: infusion 6 hours after reactivation produced no amnesia, and the memory remained retrievable shortly after the manipulation, with the impairment appearing only at later testing (Elsey et al., 2018). This is the finding that puts restabilization in the substrate rather than in the interpretation.
Human evidence exists but is correlational and imaging-based. Agren et al. (2012) conditioned 22 participants to a visual cue paired with shock, reactivated the memory the following day with a single 2-minute presentation of the cue, and then gave extinction training either 10 minutes later, inside the presumed reconsolidation window, or 6 hours later, outside it. Two days after that, functional imaging during renewal showed significantly greater bilateral basolateral amygdala activity in the 6 hours group, and in that group alone amygdala activity predicted the return of fear on a subsequent reinstatement test and covaried with fear recall during the previous day's extinction. The same amygdala seed showed stronger coupling with insula, hippocampus, and midline anterior cingulate in the 6 hours group. The authors argue against the competing account, that extinction merely suppressed the memory through prefrontal inhibition, on the grounds that ventromedial prefrontal activity did not correlate negatively with return of fear in either group (Z-scores below 1).
The two lines of evidence are not doing the same work and should not be read as one. The rodent result is an intervention on a molecular process with memory loss as the outcome. The human result is a difference in blood-oxygen-level-dependent signal between two extinction timings, with no group that received neither reactivation nor extinction, so the inference from imaging to destabilization runs through the behavioural manipulation rather than around it. Elsey et al. (2018) make the general form of this point: reconsolidation cannot be directly observed in humans and is inferred from later changes in memory expression, so behavioural and imaging findings can be consistent with the process without establishing it.
AI at a Glance
In humans, retrieving a consolidated memory returns it to an unstable state in which it can be weakened, strengthened, or updated with information present at the time of retrieval, and it must be rebuilt over a period of hours in order to persist (Nader et al., 2000; Hupbach et al., 2007). On the AI side the closest match is a memory layer that flags a just-retrieved record as modifiable for a fixed window and blends contradicting information into it rather than storing a second record, a component that has been built and ablated but has never been put in front of an evaluation containing the contradictions it responds to (Kerestecioglu et al., 2026).
- Memory layer: A store of text or structured records kept outside a model's parameters and outside its current input, written and read by surrounding software rather than by the model's own learning (Kerestecioglu et al., 2026).
Reconsolidation in AI
How Far It Has Been Measured
Measurement in AI: Indirectly Measured
AI Memory Performance: Comparison Not Possible
Where It Lives in the System
What Has Been Measured
- In training (the model's built-in knowledge, stored in its weights), knowledge-editing methods rewrite individual facts well but fail to carry the change through: on GPT-J, edited facts were recalled 88% (ROME) to 96% (MEMIT) of the time, yet accuracy on multi-hop questions that depend on them fell from 40.5% to about 7% (Zhong et al., 2023).
- Within the current conversation, when facts in the context window (what the model can see) conflict with what it learned in training, models accept coherent, convincing new evidence but also show confirmation bias toward their stored knowledge (Xie et al., 2024).
- A workshop paper adapted the human A-B, A-C proactive interference design to the conversation: as a value was repeatedly updated, models' accuracy at reporting the latest value declined log-linearly toward zero because they retrieved overwritten values; larger models held up better, but no human participants were tested (Wang & Sun, 2025).
- Saved-memory systems do update on retrieval in a loose sense: Mem0 retrieves stored memories similar to each new fact and has the language model choose to add, update, delete, or leave them, though this is an engineering design rather than a tested analogue of reconsolidation (Chhikara et al., 2025).
- A 2025 benchmark preprint tested whether agents overwrite older facts with newer contradicting ones: on single-fact questions scores ranged from 7% (Zep) to 78% (GPT-5-mini), and on multi-step questions every system scored 28% or lower, with most at 2-7%; no human baseline was reported (Hu et al., 2025).
- Comparison is not possible because there is no common scale: no study located runs AI on a reconsolidation paradigm, and human studies measure a change that depends on retrieval, against a no-reactivation control (e.g., Hedges g), whereas AI studies only score whether an update succeeded and report no human results (Pigeon et al., 2022).
Where a Mechanism Can Live
04 Product
The whole configuration, and the only thing a benchmark observes directly.
01
Weights
Across deployments
02
Context Window
Within a session
03
Memory Layer
Across sessions
Which of the three a result belongs to is inferred from it, not seen in it.
Read more
Amygdala
Subcortical in the medial temporal lobe
An almond-shaped cluster of a dozen distinct nuclei in the anterior medial temporal lobe, not a single functional organ.
Grey matter in front of the hippocampus, conventionally divided into basolateral, central, medial and cortical groups whose cell types, connections and outputs differ enough that treating them as one structure obscures most of what is known. The "fear center" label is wrong on two counts: the nuclei do different things, and the responses they are implicated in are not specific to fear. Recordings and lesions support involvement in learning about biologically significant events of either valence, and whether the common factor is valence, arousal, or salience remains genuinely open.
Anterior cingulate cortex
Cortex in the cingulate cortex
The front section of the cingulate, curving around the genu of the corpus callosum, with subgenual, pregenual, and dorsal parts.
The anterior cingulate cortex wraps the front of the corpus callosum, occupying Brodmann areas 24, 32 and 33, and on some maps 25. It is connected with the amygdala, the autonomic brainstem, the ventral striatum and lateral prefrontal cortex, and in humans and great apes it contains von Economo neurons. Its function is one of the most openly contested in the field: the same dorsal tissue has been proposed as a conflict monitor, an error detector, a tracker of the expected value of control, an effort or foraging-cost signal, and part of the pain network. The proposals are not variants of one claim; they make different predictions and coexist.
Basal ganglia
Subcortical
A set of interconnected subcortical nuclei (striatum, pallidum, subthalamic nucleus, substantia nigra) forming loops with cortex and thalamus.
Not a system defined by adjacency but by circuitry: cortex projects to striatum, striatum to pallidum and substantia nigra, and these back to cortex through thalamus, in several parallel loops that stay segregated from motor through associative to limbic territory. The "direct and indirect pathway" model organized the field and explains a good deal of Parkinson's and Huntington's phenomenology, but its predictions about firing rates have not held cleanly, and much current work treats pattern and timing as well as rate as what the loops carry.
Basolateral amygdala
Nucleus in the amygdala
The cortex-like input division of the amygdala, where sensory and contextual signals converge with information about outcomes.
The lateral, basal and accessory basal nuclei taken together, built from glutamatergic principal cells that look and behave much more like cortical pyramidal neurons than like striatal cells. It receives sensory input from thalamus and cortex, reciprocal connections with prefrontal cortex and hippocampus, and projects to the central nucleus, nucleus accumbens and back to cortex. This is where associations between cues and outcomes are most often localized, in aversive and appetitive conditioning alike; the appetitive half is routinely dropped in summaries that call it a fear structure.
Hippocampus
Cortex in the hippocampal formation
A curved allocortical structure in the medial temporal lobe whose bilateral damage reliably produces dense anterograde amnesia in humans.
A paired, seahorse-shaped fold of three-layered cortex running front to back inside the temporal lobe, connected to the rest of the brain mainly through entorhinal cortex and the fornix. What it does is genuinely contested and has been for fifty years: the human lesion literature reads it as necessary for forming new declarative memories, the rodent recording literature as a map of space, and a third position treats both as instances of binding arbitrary elements into relations. The accounts differ on which species and which task they explain best, and the question of which is right remains open.
Insula
Cortex in the cerebral hemisphere
A triangular expanse of cortex buried inside the Sylvian fissure, covered over by the frontal, parietal and temporal opercula.
The insula is the only cortical lobe with no exposed surface: it sits at the bottom of the Sylvian fissure, hidden by the opercula that grow over it during development, and is divided by its central sulcus into a larger anterior and smaller posterior part. Its architecture grades from agranular at the front to granular at the back. It receives visceral, thermal, nociceptive and gustatory input. Popular accounts name it the seat of interoception, disgust, empathy, or consciousness in turn; the region appears in a large share of published fMRI contrasts, which makes reverse inference from insular activity to any one of those especially unsafe.
Prefrontal cortex
Cortex in the frontal lobe
The frontal cortex in front of the motor areas, defined classically by its connections with the mediodorsal thalamus rather than by any one function.
The prefrontal cortex occupies the front of the frontal lobe, anterior to premotor and motor cortex, and accounts for roughly a sixth to a quarter of human cortex depending on where the posterior border is drawn. Three competing definitions are in use (projection from the mediodorsal thalamus, the presence of a granular layer IV, and the absence of movement on electrical stimulation), and they do not enclose the same tissue, which is why rodent-to-primate homology is still argued over. It is reliably engaged by tasks with delays, rules, and conflicting options; what it contributes to those tasks is the open question, not whether it participates.
Elsey et al. (2018)
Human memory reconsolidation: A guiding framework and critical review of the evidence.
In
Psychological Bulletin, 144(8), 797-848
Abstract
Research in nonhuman animals suggests that reactivation can induce a transient, unstable state in a previously consolidated memory, during which the memory can be disrupted or modified, necessitating a process of restabilization in order to persist. Such findings have sparked a wave of interest into whether this phenomenon, known as reconsolidation, occurs in humans. Translating research from animal models to human experiments and even to clinical interventions is an exciting prospect, but amid this excitement, relatively little work has critically evaluated and synthesized existing research regarding human memory reconsolidation. In this review, we formalize a framework for evaluating...
Find it
Hu et al. (2025)
Evaluating Memory in LLM Agents via Incremental Multi-Turn Interactions
In
arXiv
Abstract
Recent benchmarks for Large Language Model (LLM) agents primarily focus on evaluating reasoning, planning, and execution capabilities, while another critical component-memory, encompassing how agents memorize, update, and retrieve long-term information-is under-evaluated due to the lack of benchmarks. We term agents with memory mechanisms as memory agents. In this paper, based on classic theories from memory science and cognitive science, we identify four core competencies essential for memory agents: accurate retrieval, test-time learning, long-range understanding, and selective forgetting. Existing benchmarks either rely on limited context lengths or are tailored for static,...
Find it
Kerestecioglu et al. (2026)
Human-Inspired Memory Architecture for LLM Agents
In
arXiv
Abstract
Current LLM agents lack principled mechanisms for managing persistent memory across long interaction horizons. We present a biologically-grounded memory architecture comprising six cognitive mechanisms: (1) sleep-phase consolidation, (2) interference-based forgetting, (3) engram maturation, (4) reconsolidation upon retrieval, (5) entity knowledge graphs, and (6) hybrid multi-cue retrieval. Each mechanism addresses a specific failure mode of naive memory accumulation. We introduce a synthetic calibration methodology that derives all pipeline thresholds without benchmark data exposure, eliminating a common source of evaluation leakage. We evaluate on two benchmarks. First, a VSCode...
Find it
Nader et al. (2000)
Fear memories require protein synthesis in the amygdala for reconsolidation after retrieval
In
Nature, 406(6797), 722-726
Abstract
‘New’ memories are initially labile and sensitive to disruption before being consolidated into stable long-term memories1,2,3,4,5. Much evidence indicates that this consolidation involves the synthesis of new proteins in neurons6,7,8,9. The lateral and basal nuclei of the amygdala (LBA) are believed to be a site of memory storage in fear learning10. Infusion of the protein synthesis inhibitor anisomycin into the LBA shortly after training prevents consolidation of fear memories11. Here we show that consolidated fear memories, when reactivated during retrieval, return to a labile state in which infusion of anisomycin shortly after memory reactivation produces amnesia on later tests,...
Find it
Pigeon et al. (2022)
Impairing memory reconsolidation with propranolol in healthy and clinical samples: a meta-analysis
In
Journal of Psychiatry and Neuroscience, 47(2), E109-E122
Abstract
Background Reconsolidation impairment using propranolol is a novel intervention for mental disorders with an emotional memory at their core. In this systematic review and meta-analysis, we examined the evidence for this intervention in healthy and clinical adult samples. Methods We searched 8 databases for randomized, double-blind studies that involved at least 1 propranolol group and 1 placebo group. We conducted a meta-analysis of 14 studies ( n = 478) in healthy adults and 12 studies in clinical samples ( n = 446). Results Compared to placebo, reconsolidation impairment under propranolol resulted in reduced recall of aversive material and cue-elicited conditioned emotional responses in...
Find it
Wang & Sun (2025)
Unable to Forget: Proactive Interference Reveals Working Memory Limits in LLMs Beyond Context Length
In
arXiv
Abstract
Information retrieval in Large Language Models (LLMs) is increasingly recognized as intertwined with generation capabilities rather than mere lookup. While longer contexts are often assumed to improve retrieval, the effects of intra-context interference remain understudied. To address this, we adapt the proactive interference (PI) paradigm from cognitive science, where earlier information disrupts recall of newer updates. In humans, susceptibility to such interference is inversely linked to working memory capacity. We introduce PI-LLM, an evaluation that sequentially streams semantically related key-value updates and queries only the final values. Although these final values are clearly...
Find it
Xie et al. (2024)
Adaptive Chameleon or Stubborn Sloth: Revealing the Behavior of Large Language Models in Knowledge Conflicts
In
arXiv
Abstract
By providing external information to large language models (LLMs), tool augmentation (including retrieval augmentation) has emerged as a promising solution for addressing the limitations of LLMs' static parametric memory. However, how receptive are LLMs to such external evidence, especially when the evidence conflicts with their parametric memory? We present the first comprehensive and controlled investigation into the behavior of LLMs when encountering knowledge conflicts. We propose a systematic framework to elicit high-quality parametric memory from LLMs and construct the corresponding counter-memory, which enables us to conduct a series of controlled experiments. Our investigation...
Find it