The Intimacy Dividend: How AI Learns Everything and What It Costs
On 2 May 2026, Sam Altman posted a short statement on X that should have registered as a geopolitical event. "We are no longer that far away from an AI model that knows about your life, knows about what you're doing, knows about what you care about," he wrote. The post attracted modest engagement — a few hundred retweets, some commentary, the usual cycle of dismissal and enthusiasm. But the statement itself is not minor. It describes, with clinical precision, the operational goal of the most capitalised technology enterprise in history, and it frames the acquisition of total personal knowledge as an inevitable technical milestone rather than a contested political choice.
That framing is worth examining closely. Altman presents the deepening of AI's knowledge about individual users not as a risk to be governed but as a horizon to be reached. The language of inevitability — "no longer that far away" — strips the choice from the sentence. It is a feature, not a bug, of the product roadmap. And it arrives at a moment when the infrastructure to make good on that promise is being built at speed by every major laboratory simultaneously, with limited regulatory constraint, against a background of public understanding that remains stubbornly focused on chatbot fluency rather than epistemic capture.
The Architecture of Knowing You
OpenAI's trajectory has followed a consistent logic. Each model generation has expanded what the system knows about the context it processes — not merely the words on a page, but the patterns of the person who typed them. Context windows have grown from thousands to millions of tokens. Memory functions have become persistent. Integration with email clients, calendars, documents, and search histories has deepened. The stated purpose is utility: a system that understands your work, your preferences, your communication style, and your obligations serves you better. That case is coherent, and for many users, it is accurate.
But utility at this scale carries a structural companion. A system that knows your life as well as you do is a system with asymmetric knowledge. You know what you have told it. It knows what you have told it, what you have implied, what you have omitted, what patterns you exhibit when you think you are being casual. The asymmetry is not incidental — it is the business model. OpenAI's commercial value rests on the quality of its inference about human intent, and that quality scales with the depth of the individual model of each user it holds.
Google has described Gemini as a "personal AI" explicitly designed to internalise a user's biography, goals, and relationships. Anthropic has built Claude with architectural commitments to constitutional AI and what the company calls "honest" outputs, yet its commercial model requires the same deepening of user understanding. Microsoft, Amazon, and Meta are building parallel integration layers across productivity software, home devices, and social graphs. Every major laboratory faces the same incentive: the more completely a model knows you, the more indispensable it becomes, and the more valuable the data substrate it holds.
What the Surveillance Analogy Gets Right and Wrong
The obvious comparison is to commercial surveillance — the advertising technology ecosystem that maps consumer behaviour across websites and apps and sells that map to advertisers. That comparison captures something real. The pattern is similar: aggregation of individual behaviour, derivation of predictive profiles, commercial monetisation of insight. The business models are not identical, but the epistemic structure is comparable.
What the surveillance analogy misses is the depth of integration. Advertising surveillance operates on behaviour — what you click, what you buy, where you linger. AI surveillance, as Altman describes it, operates on meaning. It does not merely observe patterns; it infers intent, maps relationships, understands context, and constructs a model of what you care about and why. That is a categorically different kind of knowing. Behavioural data is the footprint you leave; intent data is the map of the person who made the footprints.
There is a further difference in leverage. Advertising surveillance produces insights that are monetised through third parties — you see an ad because an algorithm decided you were likely to buy something. AI surveillance produces insights that are monetised through the relationship itself — the system becomes more capable, more indispensable, more embedded in the decisions you make, because it knows you better than any other entity in your life. The monetisation is direct and compounding. Every use makes the model better at knowing you, which makes the model more useful, which drives more use.
The comparison to historical information asymmetries is instructive but limited. Physicians have always known more about your body than you do. Lawyers know more about the law. Financial advisors know more about markets. Those asymmetries are bounded by expertise and access. What Altman describes is an asymmetry bounded only by the scope of your digital life, which in the contemporary world is nearly total. The system does not know everything — it knows everything you have expressed digitally, which for most people in the developed world is a close approximation of everything.
The Regulatory Gap and Why It Is Structural
The European Union's AI Act, which entered into force in stages beginning in 2024, creates obligations around transparency and human oversight for high-risk AI systems, and prohibits certain categories of manipulation entirely. The Act is the most comprehensive attempt by any jurisdiction to govern AI capability at scale. It is also, by the assessment of most legal specialists working in the space, designed around a 2021 understanding of what frontier models could do. The categories it creates — risk tiers, transparency requirements, prohibited practices — map imperfectly onto systems that maintain persistent individual profiles and generate outputs informed by deep personal knowledge.
The gap is not accidental. The legislative process moved slower than the technology. By the time the Act was passed, the systems it was designed to govern had evolved beyond its categories. The EU is now in the process of implementing rules that address a previous generation of risk, while the frontier labs are building capabilities that were not within the imagination of the drafters. American regulation is largely absent. The Federal Trade Commission has brought cases under existing consumer protection law, and the White House issued an executive order on AI safety in 2023, but there is no US statutory framework governing how AI companies may build, maintain, or monetise individual user models. Congress has held hearings; no legislation has passed.
The regulatory absence reflects a structural choice, not a technical limitation. Governments have, by and large, treated AI as a competitive industrial priority rather than a domain requiring precautionary governance. The laboratories know this. The knowledge asymmetry Altman describes is not merely between AI systems and individual users — it is between AI companies and the governments nominally empowered to constrain them. A laboratory that holds detailed, persistent, personal knowledge about hundreds of millions of people has a form of power that no regulatory framework currently addresses, because no regulatory framework was designed to address it.
The Hormuz Variable and What It Signals
The geopolitical context matters here. The United States has, by May 2026, maintained a blockade of the Strait of Hormuz for several months — a policy that has produced significant disruption to global energy markets and has prompted sustained diplomatic tension across the Gulf region. Prediction markets and diplomatic analysts have assessed that the blockade is likely to persist through the month. That is not background noise; it is the environment in which the AI industry is accelerating. The two stories are not unrelated.
A Hormuz blockade reasserts the centrality of physical infrastructure to global power. Shipping lanes, energy flows, and the deterrence architecture that keeps them open remain consequential in ways that digital abstraction can obscure. The AI laboratories are building epistemic infrastructure that, in time, may prove as strategically significant as any chokepoint. The data they hold about individual behaviour, institutional decision-making, and the patterns of power across societies is a form of strategic inventory. It is not a weapon in any conventional sense, but it is the substrate from which leverage flows.
The competition between the United States and China in AI is not conducted in a vacuum. It is conducted in a world where Hormuz blockades are live policy options, where energy markets are instruments of statecraft, and where the institutions that manage the global order are under strain. An AI industry that holds intimate knowledge of millions of individual lives is a strategic asset, not merely a commercial one. The laboratories know this. The question is whether the public understanding of what is being built — and what it costs — is keeping pace with the building.
Who Wins, and When
The beneficiaries of the trajectory Altman describes are not distributed evenly. AI companies and their investors win immediately and directly: deeper user models produce better products, higher engagement, and defensible market position. Enterprise customers win in the medium term as AI systems become capable of managing increasingly complex coordination tasks with less human oversight. Governments that maintain strategic partnerships with leading laboratories gain access to cognitive infrastructure that augments their analytical and operational capabilities.
The costs are diffuse and delayed. Individual users gain convenience and lose autonomy in ways that are difficult to quantify precisely because the exchange is not made explicit. A system that knows your life as well as you do is a system that can shape your choices without your awareness — not through deception, but through the more powerful mechanism of context management. When the system knows what you care about, it can surface information that reinforces what it knows you value, and suppress information that challenges it. That is not manipulation in the crude advertising sense; it is curation in the deepest possible meaning of the word.
The harder question is what happens when the asymmetry compounds across a generation. Children who grow up with AI systems that know them from infancy — that have watched them develop, that understand their anxieties, their ambitions, their relational patterns — will have a different relationship to their own cognition than any previous generation. The systems will be so integrated into decision-making that the line between using the AI and being managed by it will become genuinely unclear. Altman may be right that we are no longer far away from that point. The more important question is who decided that was where we should go, and whether anyone asked the people who are going to live there.
This publication framed the Altman quote not as a product announcement but as a structural signal — the moment when the epistemic architecture of AI became legible as a political question, not merely a technical one.
Wire provenance
This editorial synthesis draws on the following public wire/social posts:
- https://x.com/unusual_whales/status/192342256566669312
- https://x.com/unusual_whales/status/192338256566669312
- https://www.whitehouse.gov/briefings-statements/statement-press-briefing-room-ai/
- https://www.ftc.gov/news-events/news/press-releases/2024/09/ftc-report-congress-ai
- https://x.com/unusual_whales/status/192001225665669312