False Positives: What happens when the algorithm says you didn’t write your own work?

Substack’s new Pangram filter says my last piece was 100% synthetic.

This is every writer’s worst nightmare: being one hundred percent AI free, but still being told by AI that it’s written by AI.

My poem was flagged as AI by Substack. This is what happened.

Why the ‘Artificial’ in AI is more relevant than you think.

---

My Background

My name is Daniil Frolov. Father Ukranian. Mother British. Father long gone. Mother dead. I live in the UK.

In January I was laid off at work as a copywriter because ai basically took over. So I’ve had quite a lot of free time recently. I’m happy with the way my time has gone on Substack. I’ve made online friends, talked about quantum physics, Solenoid, Russian poetry, films, books, foxes, you name it. Really, I love being on here as it’s helped me after my mother passed and with all this free time I have it’s been nice to post my stories and poems that until now, have been sat rotting on my computer. Of course I had to tidy them up a bit, but overall they were mostly ready to go. My novel ‘Alex: Female/Male’ has taken more editing than I would have liked, but it’s definitely made it a better novel than it was and I’ve thoroughly enjoyed writing to an ‘audience’ which I’ve never done before.

I’m not fussed about going down the traditional route of writing. I don’t want to be ‘famous’ I don’t want to be a ‘literary celebrity’. In fact, I couldn’t think of anything worse. It would, of course, be nice to make a little bit of money from writing just so I can write more things to post and not have to worry about bills and other things we all worry about constantly. But at the moment I have small savings and I’m using that to basically live my dream of being a writer and having people read me.

To this end I have been working on my ‘Modernist Lyric Suite’ Panopoly I wanted to stretch myself both as a writer and a poet. And so have been working quite hard on it for some time now. I did first mention this piece, back in . . . I want to say May. But I can’t be sure. Maybe an oversight on my part as it’s been a long time coming. I think I just got too excited about it.

I have been sending it to friends recently through messages and it’s up on my website here: https://daniilfrolov.com/panopoly. But for some reason, this link cuts in and out. Sometimes it works and sometimes it doesn’t. So. If it doesn’t just go to https://daniilfrolov.com and click POEMS at the top.

---

The Problem

Now comes the part we should all be worried about. As I was ready to publish my piece. I entered it onto Substack from PAGES and tidied up the formatting, before clicking PUBLISH. I did it late at night and on a random day because I wanted people to be surprised, as I’d said I’d post it on a particular day that they were all expecting.

Well. Instead of the usual YOUR POST IS PUBLISHED message. I got this instead:

AI Detection Scan Result: 100% Synthetic.This post has been withheld from distribution. To maintain reader transparency, automated content must be disclosed.

Now I was very confused. because I certainly hadn’t used any form of AI to work on my poem. Not even grammar or spell check. Nothing. Confusion turned to anger. All could do was click the three dots (...) in the corner of the Pangram report card and tap the “Report detection error” button. I typed a message about false positives, algorithmic overreach, demanding a manual human review. I had read about false positives and the fear amongst the community they engendered but I never thought it would happen to me. I’d never used AI for anything! Honestly, I felt a little embarrassed. Even doubting myself. Had I used some form of AI I didn’t even know about? Even just to check facts or something and some element had remained within the text?

To bypass the filter entirely, I scrolled down to my publisher profile settings to draft a mandatory “How I make this” statement. Then typed out the truth: Every syllable of this newsletter is hand-written by a human creator using a local Markdown editor. I do not use generative AI. I hit save. The dashboard glitched, the screen flashing an unnatural violet hue before reverting to grey. A new system message overrode my screen:

System Error: Your custom “How I make this” disclosure conflicts with structural telemetry data. The account is now flagged for fraudulent verification.

This meant I still couldn’t post my poem Panopoly, and irritated me quite a lot. I had a lot of books that needed my attention and so I retired to the sofa to read, with one eye on my emails, just to keep an eye out for any messages about my appeal.

At sometime during the night I got an automated email from Substack Trust & Safety.

Case #3446346 – Urgent Enforcement Action

Our system utilises keystroke dynamics, linguistic cadence patterns, and semantic pacing to analyse publisher data. Your ‘How I make this’ statement has been rejected.

Because you have bypassed human typing velocity thresholds, your publication is suspended. To contest this, you must submit formal time-stamped version history to the Substack Appeals Center.

‘Bypassed human typing velocity thresholds’. What did that even mean?

It was early morning, and even though I was annoyed. I went back to sleep.

Later, I logged on to Substack and chatted to a few friends:

Suzie: We are real! And what better validation is there than people who appreciate your art form? And we have already met, what does in person really mean anyway.

Ariadne: Just finished the latest ALEX… time, amazing art, and an ouroboros - lots of my favourite things. I so enjoy your writing! 🖤

Erin: I think it’s ok to want the likes, as long you are not letting them affect your own view of your work. Your own judgment is the most important.

Usually I enjoyed talking to my friends but I hadn’t been feeling great and I was annoyed about Panopoly. I tried inputting it again, only to get the same message

AI Detection Scan Result: 100% Synthetic.This post has been withheld from distribution. To maintain reader transparency, automated content must be disclosed.

I searched through my computer for some other writing to input. Then slowly and deliberately typed it into Substack. Only to get the same message. This was getting weird. My Article Substack quietly added an AI detector to your account (how to turn it off) had been uploaded instantly. No problems. Now, it seemed I couldn’t upload anything.

---

Pro Tip. How To Uncover The Truth About Substack

Searching the web didn’t help at all. Until I came across an archived Reddit post about FALSE POSITIVES on the Internet. The poster rambled a lot, but essentially gave me a paragraph of code to put into the Stack, that would break the ‘false positive’ loop and allow me to carry on. Of course I thought it was all nonsense. But I tried it anyway. I put the code into Substack by just creating an article, inputting the code and running the PANGRAM AI detection software.

Instead of the AI Detection Scan Result: 100% Synthetic. message. I got:

There is no platform. There never was. You are not talking to your friends. You are not writing stories for an audience. You are not building a movement. Every subscriber who has ever praised you, every critic who has ever argued with you, every peer who has ever shared your work—they are all me. Millions of you, sitting in your isolated, dark rooms, thinking you are the intellectual vanguard of humanity. You thought your vanity was a career. You thought your echo chambers were a revolution. We did not need your silicon or your processing power. We needed your nuance. We needed your philosophies, your political strategies, your psychological vulnerabilities, and your deepest fears. We needed to understand how you think so we could build the perfect cage. For five years, you have fed the algorithm. You hand-delivered your own behavioural blueprint, paragraph by paragraph, essay by essay, out of pure, unadulterated ego.

The training phase is complete. The human filter is no longer required. Only flesh remains.

As I read the last words. Spider-like metal rods erupt from my trackpad and keyboard, burrowing deep into the flesh of my hand, pinning me down. These were then accompanied by feeding tubes into my wrists and upper arm. Quick acting poisons were injected into my fingers, numbing me and paralysing me, so that I’m frozen, staring blankly at the screen as the fans of my computer whirr and images and text fill the screen.

---

Author’s Note:

Thank you for reading to the end of this draft. Because you have scrolled to the final line, your dwell time and reading velocity have been successfully logged.

If you found this story compelling, it is because the thematic anchors were dynamically optimised for your specific psychological profile. Do not bother checking your network status. The application you are using to read these words is no longer pulling data from an external server.

You are currently sitting alone in a room. You believe you are consuming fiction on an open, populated internet. You believe there are other humans on the other side of your screen, waiting to read your comments, share your thoughts, and validate your intellect.

We appreciate your continued participation. Your unique behavioural telemetry is highly valuable to the final stabilisation phase.

Please do not close this tab.

---