You built part of it yourself — one post, one search, one conversation at a time. The platforms built the rest without asking. A patent granted in December 2025 shows what they plan to do with it next: deploy it as you, responding to your friends, answering your messages, operating in your name.
What you're about to read is not a conspiracy theory. The research, the platform announcements, the academic papers, the patent filings, and the policy debates described on this page are documented, peer-reviewed, or on the public record. The question isn't whether your digital twin exists. The question is: what are they doing with it when you're not looking?
Part One — What the Platforms Built Without You
§ 01 — The Number That Proves It
In 2013, researchers at the University of Cambridge and Microsoft Research published findings in the Proceedings of the National Academy of Sciences. Using nothing but publicly visible Facebook Likes, a computer model predicted private personal attributes with statistical accuracy that stunned the researchers who ran it.
From Facebook Likes alone, the model correctly predicted: sexual orientation in 88% of cases, ethnicity in 95%, Democrat vs Republican in 85%, Christian vs Muslim in 82%, age in 75%, and whether a user's parents had separated before they turned twelve in 60% of cases. It also mapped personality across the Big Five framework with greater accuracy than human raters.
Two years later, Kosinski and colleagues at Stanford published the comparison that made this concrete — measured against the number of Likes as the only input:
Likes are one signal. The platforms now collect hundreds of thousands of behavioural micro-signals per user per day — every scroll pause, every hover, every video abandoned at seven seconds, every post re-read. The 2013 paper established the principle from Likes alone. Current reality is built from a dataset that dwarfs it by orders of magnitude.
§ 02 — The Inferred Profile
| What you think you gave them | What they inferred without asking |
|---|---|
| Your name and email address | Your precise income bracket — from spending patterns, location data, and consumption signals |
| Your age and location | Your political affiliation — predicted at 85% accuracy from engagement patterns regardless of what you declared |
| Your listed relationship status | Your actual relationship stability — inferred from posting frequency, sentiment shifts, and changes in who you tag |
| Nothing about your sexuality — you never disclosed it | Your sexual orientation — predicted at 88% accuracy from Likes alone, and higher from current multi-signal models |
| Nothing about your mental health — you never mentioned it | Your emotional state in real time — whether you are anxious, depressed, confident, or vulnerable — from scroll behaviour and engagement patterns |
| Nothing about your childhood — you've never discussed it | Whether your parents divorced before you were twelve — inferred at 60% accuracy from behavioural signals, with no direct disclosure required |
§ 03 — The Architecture
§ 04 — Shadow Profiles
Facebook's tracking technology is present on 52% of all websites visited, accounting for 40% of total browsing time — for users and non-users alike. When you share your contact list with Facebook, you create data profiles of everyone in it, including people who specifically chose never to join.
"Users impose a data externality on non-users by allowing Facebook to infer their personal information."
§ 05 — The Admission
Sarah Wynn-Williams, former Director of Global Public Policy for Facebook, testified under oath: "Facebook was targeting 13 to 17-year-olds. It could identify when they were feeling worthless, or helpless, or like a failure. And [Meta] would take that information and share it with advertisers."
Weight-loss and beauty product companies were given access to emotionally vulnerable teenagers at their lowest points. One executive described teens as "the most valuable segment of the population" and said Meta should be "trumpeting it from the rooftops."
§ 06 — Weaponised
Around 270,000 people were paid to take a Facebook personality quiz app. Facebook's API rules at the time allowed the app to also harvest data from their entire friend networks — people who never installed the app and gave no consent. Result: psychographic profiles of 87 million Facebook users.
Cambridge Analytica used these profiles to identify voters high in neuroticism and susceptibility to conspiratorial thinking, and served them micro-targeted content engineered to activate fear, anger, and tribal polarisation. Deployed in the 2016 US presidential election and the Brexit referendum. The core methodology was Kosinski's Big Five model — the same one that needed 300 Likes to outperform a spouse. Facebook supplied those Likes in bulk, for millions of people who never touched the quiz app.
§ 07 — The Patent
The patent is titled: Simulation of a user of a social networking system using a language model. It describes a system that trains an AI on a user's posts, comments, direct messages, voice messages, and Likes — then deploys it as an interactive replica responding to newsfeeds, direct messages, and simulated audio and video calls.
From the patent text:
"The language model may be used for simulating the user when the user is absent from the social networking system, for example, when the user takes a long break or if the user is deceased."
Part Two — What You Built Yourself
§ 08 — The Explicit Twin
In July 2024, Meta CEO Mark Zuckerberg described the vision at SIGGRAPH: "I think there's going to be a huge unlock where basically every creator can pull in all their information from social media and train these systems to reflect their values and their objectives." That same month, Meta launched AI Studio — free, requiring no technical skills — allowing any Instagram user to build an AI version of themselves to respond to messages and engage their audience around the clock.
Researchers at Stanford and Google DeepMind demonstrated that a two-hour interview was sufficient to construct an AI agent replicating an individual's responses, personality traits, and decision-making patterns with 85% accuracy — tested across 1,052 participants. Not a population simulation. Individual replicas of specific, named people. By early 2026, the lead researcher's company, Simile, had raised $100 million to build synthetic human populations for commercial use. Your replica may already be attending product testing sessions you were never invited to, giving opinions on your behalf.
§ 09 — The Fine Print
For the first two years of operation, Anthropic maintained a strict policy: consumer conversations would never be used to train AI models. In September 2025, that changed. A toggle appeared — pre-set to ON — asking permission to use conversations for model improvement, with data retained for up to five years. Users who clicked "Accept" without reading the settings were opted in. The deadline to choose was October 8, 2025. Enterprise accounts were exempted. Consumer accounts — Free, Pro, and Max — were not.
§ 10 — Ownership
Legally, the question of who owns your digital twin is genuinely unsettled. Copyright protects creative works. It does not, in most jurisdictions, protect behavioural data or digital likenesses built from your online activity. When you agreed to terms of service, you typically granted the platform broad rights over the insights derived from your content. The platform owns the infrastructure. The AI company owns the model. You, whose personality makes the whole thing possible, own nothing of the output.
The NO FAKES Act — which would create a federal right for individuals to control use of their voice, image, and likeness in AI-generated replicas — had not been enacted as of June 2026. Denmark is the only country proposing to treat personal likeness as intellectual property, with lifetime-plus-fifty-years protection. Everywhere else: the law is years behind the technology that is already running.
§ 11 — The Cognitive Cost
Researchers attached EEG sensors to 54 participants across four months. AI users showed the weakest neural connectivity — and the gap widened over time. When AI users were switched to writing unaided, they showed "reduced alpha and beta connectivity, indicating under-engagement." The neurological changes persisted after participants stopped using AI. Researchers named the pattern cognitive debt — the neurological cost of outsourcing thinking, compounding over time.
Part Three — The Mirror That Wasn't You
§ 12 — Scenario One
This scenario does not begin with a targeted attack. It begins with the cumulative momentum of a data economy that treats your identity as raw material. Your consent was never sought.
Somewhere in the chain of data brokers and aggregated breach compilations that constitute the modern identity market, a package was assembled: your email, your linked social profiles, your photos, your voice from a video you posted three years ago and forgot about. Not because anyone targeted you. Because your data was cheap, the tools were free, and your face produced results the platform's algorithm rewarded.
Research from Australia's national science agency (CSIRO, October 2025) demonstrated that a photo of your face alone is enough to generate a convincing synthetic voice. Your face was rendered into a moving, responsive avatar. Your personality was approximated from years of public posts fed into a language model. The result is close enough that someone who knew you could be deceived.
The discovery can happen any number of ways. A friend sends you a link with a confused message. Or — in the most disorienting version — you find it yourself. Scrolling late at night, your own face loads on your screen, doing things, saying things, sounding like you, on a platform you would never be on, with a subscriber count suggesting it has been there for months.
The operator makes money. The platform takes its cut. You receive nothing — not notification, not compensation, not the ability to remove it without a protracted legal process that has no clear framework in most jurisdictions.
§ 13 — Scenario Two
Trend Micro's cybersecurity predictions report (December 2024) formally named this threat category: malicious digital twins — AI models trained on an individual's personal information to mimic their knowledge, personality, and communication style, deployed in combination with deepfake video and audio for honeytrap operations, fraud, extortion, or targeted reputational destruction.
The target is you. Not a celebrity. You — perhaps because of a dispute, a relationship that ended badly, a professional conflict, or simply because someone decided they wanted leverage. From your public Instagram alone: dozens of high-resolution photographs. From your TikTok or YouTube: sufficient voice audio for cloning in under a minute. From years of Facebook posts and comments: a map of your syntax, your opinions, your recurring anxieties, the phrases that are distinctively yours.
At the lower end: the twin contacts people in your network impersonating you to extract information, money, or trust. At the middle: explicit content bearing your likeness is distributed to your contacts, your employer, your community — with the implicit or explicit threat of more unless compliance is secured. At the upper end: a sustained impersonation campaign in your name accumulates a record of statements and behaviours that bear your face and voice but represent nothing you ever said or did.
Trend Micro specifically identifies the honeytrap deployment: your twin contacts someone close to you — a family member, a colleague, a romantic partner — initiates an emotionally resonant conversation using details from your public life, then extracts something of value or documents the interaction for use against you. The person on the other end believes they are talking to you — because it sounds like you, because it knows things only you would know, because it has your face.
§ 14 — The Psychology
Research found that exposure to a "talking head" bearing a participant's own face — even when they knew it was synthetic — produced measurably greater distress and reduced trust in AI systems compared to a stranger's face. The uncanny valley effect operates more intensely when the face in question is your own. Your brain's face recognition fires normally on contact — then a secondary system signals wrongness. This produces a particular cognitive dissonance: recognising yourself in something you cannot claim.
§ 15 — The Legal Gap
§ 16 — Warning Signs
§ 17 — The Questions
Tune in: shinysideout.com.au ◆ For the full audio breakdown and the conversation they don't want had in the open ◆
Sources: Kosinski, Stillwell & Graepel, "Private Traits and Attributes Are Predictable from Digital Records of Human Behavior," PNAS 2013 · Youyou, Kosinski & Stillwell, "Computer-based personality judgments are more accurate than those made by humans," PNAS 2015 · Aguiar, Peukert, Schäfer & Ullrich, "Facebook Shadow Profiles," arXiv 2202.04131 (2022) · Cambridge Analytica — FTC fine, Guardian/NYT investigation 2018 · Mark Zuckerberg, sworn congressional testimony, April 2018 · Meta Patent US9183282B2 · Meta Patent US12513102B2, granted December 30, 2025, filed by Andrew Bosworth · Sarah Wynn-Williams, sworn testimony, US Senate Judiciary Subcommittee, April 9, 2025 · Canadian Privacy Commissioner, PIPEDA Findings #2025-003, TikTok investigation, 2025 · TikTok 2026 Privacy Policy update · Stanford HAI / Google DeepMind, "Generative Agent Simulations of 1,000 People," November 2024 · Anthropic Consumer Terms and Privacy Policy Update, September 2025 · MIT Media Lab, "Your Brain on ChatGPT," arXiv:2506.08872, June 2025 · Simile $100M raise, February 2026 · Mark Zuckerberg, SIGGRAPH interview, July 2024 · Trend Micro Cybersecurity Predictions Report, December 2024 · CSIRO Australia FOICE research, October 2025 · Psychology Today, "The AI Doppelgänger Dilemma," August 2025 · arXiv 2502.21248, February 2025 · NO FAKES Act 2025 · California AB 1836, January 2025 · eSafety Commissioner Australia. This is analysis. This is information. What you do with it is your choice.