I never asked for or sat a test, but the Machine profiled me anyway.

Having had a heavy few days, I needed a bit of amusement today, so I asked Claude what it knew about me. I had asked the question before, but today seemed timely as I was, in fact, writing a paper about asking this exact question, which made the exercise less abstract and more operational.

Claude was forthcoming. It gave me what it assumed was my profession, Project Manager, not bad. It gave me my home jurisdiction, the UK, and then my approximate location, which was interesting because it contradicted the UK probably because I use a VPN, but it was nice to revisit Vancouver for a brief moment. It also gave me the project files I had loaded, my preferences for the session, and the policies governing how Anthropic was using my data. All of which is, on the published terms, nothing groundbreaking there.

Then I noticed at the bottom of the window INFJ. That’s a personality type, what was that doing there? So I asked about it, and here’s what Claude said.

Specifically, your memory entry reads: High C/D (DISC). INFJ (MBTI). Strategic/Ideation/Connectedness (StrengthsFinder). Claude did not phrase this as a guess or inference, but as a fact about me, in the same flat tone it uses when you ask it about system stuff.

I had not taken a Myers-Briggs test in this conversation. I had not, on this account, taken one in any conversation, nor had I at any point, asked the AI to assess my cognitive style. I had simply asked it to help me research and write.

The personality assessment came as a free upgrade!

I noticed that addition, and it piqued my interest. It had not been divulged, or negotiated yet here it was, clear as day on the corner of my screen, ironically enough while I was writing a paper on how AI knows things about you it perhaps shouldn’t.

I sat with that for a sec.

Then I asked where the assessment had come from. Claude is surprisingly honest when asked about its mechanisms, and detailed that the information had come from accumulated context across previous conversations, from preferences injected into the session, and from inference based on my interaction style. The assessment was helpful, it offered, in tailoring its responses to my preferences.

I thought for a second about how many times I had sworn at it, and wondered if that was an INFJ thing too!

The architecture

So it turns out that Anthropic activated memory across all free and paid Claude accounts on 2 March 2026 [1]. The company calls the mechanism ‘additive’. Claude does not store a transcript, it stores extracted facts and preferences, server-side, account-scoped [1]. You can ask the model what it remembers. What it shows you is a friendly summary, written fresh each time you ask. The actual record, the database row Anthropic holds on you, stays on the server. You see the description, you do not see the file [2].

Mozilla maps the same territory in three layers [3]. Layer one is what you typed: name, location, declared preference. Layer two is metadata: what you clicked, how long you stayed, how fast you typed and how many mistakes you made. Layer three is inferred data: things the system worked out about you by comparing layers one and two with patterns from millions of other users. Layer three is where the value is. Layer three is also, as Mozilla put it, the layer where it is pretty much impossible for you to control or even discover what the machines think about you [3].

My INFJ tag was layer three. I had not declared it or clicked on it. The machine had inferred it from the way I write and the way I argue, helpfully, in good faith, in order to serve me better.

This is, by the way, the same MBTI framework used by 88 per cent of Fortune 500 companies and a substantial chunk of the political micro-targeting industry [4] [5].

So when Claude tags me as an INFJ, it is a marketing primitive. The same four letters that are used to decide which version of a campaign message you see, which job advert is shown to you, which candidate the recruiter is told to call.

I never sat for the test

I will be honest about what I felt when this landed. I felt the small, very specific irritation of a project manager who has had a milestone added to her plan by somebody who did not consult her about it. I felt the slightly larger irritation of a writer who has had her cognitive style described to her by software whose understanding of her was assembled from material she had not chosen to disclose, and I felt the smallest, most administrative version of the irritation, which was the one I sat with longest, and which is the one that produced this piece.

I had not consented to a profile.

I had clicked Continue. I had clicked Accept. I had used the product.

Anyone who tells you these are the same thing is, with respect, not paying attention.

The methodology I have spent the last five months of documented AI work building is, in part, the discipline of asking. The methodology is called contAIn. The discipline is to periodically ask the AI what it has on you, the question is muddier and the answer changes more often than you’d think.

What is funny about today is that the discipline that catches the architecture caught the architecture, while I was writing the document about the architecture, in real time, during the writing. I am the evidence. The white paper is the receipt.

The cheese has a passport

While I was writing this, I went to put the kettle on, because that is what the British do when the surveillance architecture has just classified them. I opened the fridge. The Parmigiano Reggiano in there has a microchip in it.

Not a metaphor, a real chip, embedded in the rind. Tracked from cow to deli counter, with a serial number, and a verified provenance. The cheese has a tracker ID. I do not, yet have one, but if Blair and his bunch keep going it won’t be long.

So the cheese gets full traceability. I get a personality assessment I never asked for. The cheese is allowed to retain its mystery, I am not.

The architecture is everywhere, the architecture technically asks you nothing, yet it knows everything.

What is sitting in your account right now

If you are reading this on Medium from inside the UK or Australia, the platform you are reading on may have asked you to upload your face for biometric age verification before you could comment, or read certain posts, or use the chat function. The face goes to a company called Persona, which is the Greek word for theatrical mask. The mask company asks you to submit your real face so it can confirm you are not somebody else’s mask. Substack says the biometric data is “redacted” after seven days. Redacted is not deleted. Redacted is obscured, the data still exists.

A privacy researcher in Zurich who investigated Persona via LinkedIn’s verification system found that the platform routes data to seventeen subprocessors, all based in North America. Among them: Anthropic, OpenAI and Groqcloud, listed as handling “data extraction and analysis”. The Register confirmed the same set of subprocessors, including AWS, Google, OpenAI and Anthropic. So the face you uploaded to prove you are an adult on Substack may have passed through several AI companies, two cloud providers, and assorted analytics infrastructure. None of which you negotiated, all of which was disclosed somewhere you did not read, in language designed to discourage you from reading it.

The same architectural pattern, different vendor, same consent that is not consent.

Try it, tonight, on any AI that you use.

Ask it. What do you know about me, and what happens to my data?

The answer is not abstract. It is sitting in your account right now.

It is, as it turns out, sitting in mine, with my MBTI profile attached. I never sat for the test, the machine profiled me directly.

A note on the MBTI

For anyone reading who is not familiar with the framework: the Myers-Briggs Type Indicator was developed by Katharine Cook Briggs and her daughter Isabel Briggs Myers during the Second World War, based on Carl Jung’s 1921 work Psychological Types. The first MBTI manual was published in 1962 [4]. The assessment sorts personality into sixteen four-letter codes across four dichotomies: introversion or extraversion, sensing or intuition, thinking or feeling, judging or perceiving.

Its scientific status is contested. Wikipedia describes the MBTI as making “pseudoscientific claims” [4]. Psychologist Hans Eysenck argued that the sixteen types do not fully reflect Jung’s original theory, which used thirty-two [4]. The Myers-Briggs Company itself states the assessment “should never be used in recruitment or selection because it does not measure a person’s skills or abilities” [11]. None of which has stopped 88 per cent of Fortune 500 companies and a substantial chunk of the political micro-targeting industry from using it anyway [5].

INFJ, the four letters Claude assigned me, is nicknamed “The Advocate” or “The Counsellor.” The profile, drawn from the Myers and Briggs Foundation’s own descriptions, includes: seeks meaning and connection in ideas, relationships, and material possessions; wants to understand what motivates people and is insightful about others; conscientious and committed to firm values; develops a clear vision about how best to serve the common good; organised and decisive in implementing that vision [12].

Apparently INFJ is the rarest type in the population, accounting for roughly 1-2% [12]. So I discovered by accident via AI that I am, allegedly, in the top 2% of the population. Nobody sent a certificate or held a ceremony so it was a bit of an anticlimax, plus as they warn you AI makes mistakes, I could be something entirely different but that is not the point. The point is there was a four-letter tag sitting in the corner of a chat window, labelling me, and I had not asked for it.

If you want to take a free test yourself, 16Personalities offers one with no email or payment required. The company is NERIS Analytics, based in Cambridge. They do not sell your data, you can take the test under a pseudonym, and the only meaningful tracking is the standard Google Analytics and Hotjar cookies [13].

One thing worth noting, in the spirit of this piece: 16Personalities now offers an “AI Mentor” feature. According to their own terms of service, the AI Mentor is powered by Anthropic [14]. Yes, the same Anthropic. The architecture is recursive even on the way out.

The rarest 2% of the population, identified by a piece of software that cannot reliably tell which country I am in. Sounds about right.

Configure YOUR system. contAIn the chaos. Control YOUR outcome.

Sources

[1] Anthropic, “Claude introduces memory for teams at work,” 11 September 2025 (memory rolled out across all accounts including free tier on 2 March 2026). anthropic.com/news/memory

[2] AutomationSwitch, “memory.md and ‘Claude Memory’: What It Is,” 2 May 2026. automationswitch.com

[3] Mozilla Foundation, “What Does AI Know About Me? The Three Layers of Your Digital Profile.” mozillafoundation.org

[4] Wikipedia, “Myers-Briggs Type Indicator.” en.wikipedia.org

[5] Max Murphy, “Artificial Intelligence and Personality: Large Language Models’ Ability to Predict Personality Type,” Sage Journals, 2024. journals.sagepub.com

[6] Substack Help Center, “Why is Substack asking to verify my age?” support.substack.com

[7] Biometric Update, “Persona and Paravision power Substack’s biometric age estimation,” 16 December 2025. biometricupdate.com

[8] Doc Malik, “Substack Age Restriction,” 11 December 2025. docmalik.substack.com

[9] Aihola, “Anthropic Claude Government ID Verification,” April 2026. aihola.com

[10] Thomas Claburn, “Anthropic starts checking ID for some Claude users,” The Register, 16 April 2026. theregister.com

[11] The Myers-Briggs Company, “MBTI Assessment.” themyersbriggs.com

[12] The Myers and Briggs Foundation, “16 MBTI Personality Types Descriptions.” myersbriggs.org

[13] 16Personalities (NERIS Analytics Limited), Privacy Policy. 16personalities.com/terms/privacy

[14] 16Personalities (NERIS Analytics Limited), Website Terms of Use and Purchase. 16personalities.com/terms

This article was originally published on Medium. Full sources and references are available there.