{"id":31876,"date":"2026-11-09T10:36:17","date_gmt":"2026-11-09T09:36:17","guid":{"rendered":"https:\/\/wordlift.io\/blog\/en\/?p=31876"},"modified":"2026-11-09T10:36:22","modified_gmt":"2026-11-09T09:36:22","slug":"who-owns-an-ais-memory","status":"publish","type":"post","link":"https:\/\/wordlift.io\/blog\/en\/who-owns-an-ais-memory\/","title":{"rendered":"Who Owns an AI’s Memory?"},"content":{"rendered":"\n

Why we’re building a semantic memory infrastructure for AI-native organizations.<\/em><\/h1>\n\n\n\n

<\/p>\n\n\n\n

AI assistants are getting better at remembering us. But who owns what they remember: and how do we know that their memories are true?<\/strong><\/p>\n\n\n\n

We are rapidly moving from a world in which AI answers questions to one in which AI systems remember our preferences, participate in our work, interact with customers, and accumulate experience.<\/p>\n\n\n\n

ChatGPT remembers conversations. Claude maintains persistent memories. Both now consolidate them in the background through a process each company calls dreaming<\/em>, an idea with a long history in neuroscience and machine learning, where sleep-like phases replay and reorganize what was learned. Both approaches are advancing quickly.<\/p>\n\n\n\n

Yet the emergence of persistent AI memory raises questions that go well beyond the quality of the next answer.<\/p>\n\n\n\n

What happens when an agent remembers something incorrectly? When a business decision changes? When knowledge shared in confidence becomes available in another context? When different agents develop contradictory understandings of the same organization?<\/p>\n\n\n\n

And, perhaps most fundamentally, who controls what the system learns from the information we provide?<\/p>\n\n\n\n

At WordLift, we believe these questions are connected. We are building a semantic memory infrastructure designed to preserve organizational meaning while allowing AI systems to learn, collaborate and act.<\/p>\n\n\n\n

Our hypothesis is that trustworthy agent memory requires more than the ability to retrieve the past. It requires identity, provenance, temporal reasoning, explicit uncertainty, and governance<\/strong>.<\/p>\n\n\n\n

When company knowledge slips out of your control<\/h2>\n\n\n\n

One of the most important consequences of generative AI adoption is the movement of organizational knowledge into systems that organizations do not necessarily control.<\/p>\n\n\n\n

Employees ask AI assistants to summarize internal documents, review contracts, analyze commercial proposals, prepare strategy and interpret customer relationships.<\/p>\n\n\n\n

Often, these are precisely the materials that represent a company’s most valuable knowledge.<\/p>\n\n\n\n

Sometimes this information is deliberately disclosed to an approved enterprise service under contractual protections. Sometimes it is submitted through personal accounts, without the organization’s knowledge or authorization.<\/p>\n\n\n\n

These situations are very different, but they can appear identical to the employee: type a prompt, receive an answer.<\/p>\n\n\n\n

In May 2023, Bloomberg reported that Samsung had restricted employees’ use of generative AI tools after confidential source code was uploaded to ChatGPT. The case illustrated a new category of corporate information exposure: employees could disclose sensitive material simply by seeking assistance.<\/p>\n\n\n\n

There was also a documented ChatGPT data breach in March 2023. OpenAI disclosed that a software bug had exposed some conversation information and potentially payment-related details for a subset of subscribers.<\/p>\n\n\n\n

In Italy, the Garante per la protezione dei dati personali raised broader questions about data collection, the legal basis for model training, transparency, accuracy and the rights of data subjects.<\/p>\n\n\n\n

Its December 2024 decision imposed a \u20ac15 million sanction against OpenAI. That decision was annulled by the Tribunal of Rome on March 18, 2026, because of the allocation of regulatory competence under the GDPR. The court did not resolve the substantive allegations.<\/p>\n\n\n\n

Since September 2026, several researchers have publicly asked whether unpublished work they had discussed with AI assistants found its way into results later announced by the companies behind those assistants. Mathematicians Tristan Buckmaster of NYU and Andreas Thom of TU Dresden raised the question with OpenAI; another researcher raised a similar one about a biology discovery Anthropic attributed to Claude. The companies deny using the work. According to reports, OpenAI said it could not rule out that de-identified user data had helped improve its models; Anthropic said Claude is not trained on user conversations. None of these allegations has been established. What they have in common matters more than who turns out to be right: the people who contributed the knowledge have no way to verify where it went.<\/p>\n\n\n\n

These episodes should not be collapsed into a single accusation.<\/p>\n\n\n\n

A software breach, an employee voluntarily submitting confidential material, a provider retaining prompts, and a provider using permitted content<\/a> for model improvement are separate mechanisms.<\/p>\n\n\n\n

Nor is it accurate to claim that all ChatGPT or Claude conversations are used for training. Both providers distinguish consumer services from commercial offerings. As of November 2026, their published policies say that business and API inputs and outputs are not used to train their models by default, subject to specified permissions and exceptions. Consumer services have different settings and conditions.<\/p>\n\n\n\n

But this does not eliminate the underlying architectural problem.<\/p>\n\n\n\n

A company can lose control of its knowledge without anyone hacking its systems.<\/strong><\/p>\n\n\n\n

Knowledge can be dispersed across employee accounts, conversation histories, disconnected memory systems, external tools and multiple AI services.<\/p>\n\n\n\n

Even where contractual protections prevent model training, the organization may still lack a shared, authoritative view of what its agents remember, which claims are current, how corrections propagate, and who may reuse particular information.<\/p>\n\n\n\n

It may well be the largest leak in corporate history. No one can measure it yet, and that is part of the problem: it is a vast, largely invisible transfer of organizational context into AI workflows, one prompt at a time.<\/p>\n\n\n\n

We believe organizations should be able to benefit from increasingly capable models without surrendering stewardship of their knowledge<\/strong>.<\/p>\n\n\n\n

The illusion of remembering<\/h2>\n\n\n\n

The second problem is epistemic.<\/p>\n\n\n\n

When an AI assistant recalls something from an earlier interaction, we instinctively interpret that as knowledge.<\/p>\n\n\n\n

But a remembered statement is not necessarily a fact.<\/p>\n\n\n\n

Someone may have been mistaken. The agent may have inferred more than the evidence justified. A decision may have been reversed. A policy may have expired. A customer preference may apply only to a particular project.<\/p>\n\n\n\n

Persistent conversational memory can make all of these things sound equally authoritative.<\/p>\n\n\n\n

ChatGPT provides a particularly revealing example of how the boundary between helpful personalization and informed consent can become blurred. In a 2026 study of 2,050 memory entries from 80 ChatGPT users<\/a>, researchers found that 96% of the entries were created without an explicit instruction to remember them<\/strong>. The same study identified GDPR-defined personal data in 28% of entries and psychological inferences in 52%. These findings describe a specific participant sample, not every ChatGPT user; the absence of a per-entry request is also not proof that a person had disabled the overall memory feature. Nevertheless, they illustrate a real agency problem: the system may decide what belongs in a persistent portrait of someone before that person has approved the individual inference.<\/p>\n\n\n\n

Deletion is a second problem. OpenAI’s own memory documentation<\/a> explains that deleting a conversation does not necessarily delete the memory derived from it<\/strong>. Removing information fully may require changing the memory summary and deleting the original conversation and other sources that contain the detail. This is documented product behavior rather than evidence that every deletion request fails; it matters because the user’s expectation of erasure may not match the boundaries of the system’s different stores.<\/p>\n\n\n\n

These mechanisms are not limited to ChatGPT: any assistant that learns implicitly from conversation must distinguish the user’s general consent to personalization from permission to retain or share a particular fact. The lesson for organizational memory is to make consent, provenance, permitted use and deletion properties of the stored knowledge<\/em>, rather than relying on the assistant to infer them.<\/p>\n\n\n\n

Imagine an assistant remembering that a customer agreed to purchase a product. Was this a signed commitment, an optimistic sales forecast, a quotation from a meeting, or an inference the assistant produced?<\/p>\n\n\n\n

OpenAI’s June 2026 research on memory synthesis explicitly discusses problems of freshness, correctness and scalability. Anthropic’s November 2026 research shows that memory notes, linked representations and offline consolidation can help agents correct mistakes and reuse knowledge from earlier tasks.<\/p>\n\n\n\n

These are important contributions.<\/p>\n\n\n\n

They also clarify why memory is a difficult research problem.<\/p>\n\n\n\n

A system that can summarize its experience is not necessarily capable of establishing which parts of that experience deserve to become reliable knowledge.<\/p>\n\n\n\n

The real question is not just what an AI remembers. It is whether that memory is accurate, up to date, and supported by evidence.<\/strong><\/p>\n\n\n\n

From conversational history to semantic memory<\/h2>\n\n\n\n

Our approach starts with a distinction between an event, a claim, and a verified organizational fact.<\/p>\n\n\n\n

An event records what happened: a conversation, an observation, a document update or a decision.<\/p>\n\n\n\n

A claim states something about the world. It may be extracted from an event, inferred by an agent or asserted by a person.<\/p>\n\n\n\n

A verified or approved organizational fact has a different standing. It must be supported by appropriate evidence and governed by the rules of the organization.<\/p>\n\n\n\n

These representations should remain connected, rather than collapsing into one another.<\/p>\n\n\n\n

A claim should retain its source. Its subject should refer to a stable entity. Its temporal validity should be represented explicitly. Corrections should not silently erase the history that explains how the organization reached its current understanding.<\/p>\n\n\n\n

And an agent’s confidence should never be sufficient to promote an inference into an authoritative fact.<\/p>\n\n\n\n

WoGraph is WordLift’s own knowledge graph technology:<\/strong> the infrastructure we build to represent, store, connect and query entities, relationships and evidence using semantic-web standards. It provides the canonical knowledge foundation on which memory can be linked to stable identities rather than trapped in isolated conversations. WoGraph is a technology we develop, not a generic name for any graph database.<\/p>\n\n\n\n

This is the foundation of the memory extension we are integrating into the existing WordLift Python SDK<\/a><\/strong>.<\/p>\n\n\n\n

The planned extension will bring developer primitives for episodic memory, structured claims, scoped recall, document-like workspaces, context assembly and lifecycle operations. Its purpose is to make memory an infrastructure component rather than a feature tied to one language model.<\/p>\n\n\n\n