Convicted in Five Minutes

On the morning of 13 November 2025, an animation channel with 650,000 subscribers stopped existing. Its creator, who goes by Nani Josh, had spent years building it. Every video, by his account, was original work. YouTube's notice cited “spam and scam.” He filed an appeal, as the platform invites every terminated creator to do. The rejection arrived roughly five minutes later.

Five minutes. The channel held hundreds of videos. Watching them at normal speed would take longer than a working week. Reading the appeal, opening the disputed uploads, weighing the evidence, and reaching a considered judgement about whether a years-long body of work was fraudulent: no human being did any of that in five minutes, because no human being could. The verdict had the texture of something a machine produces, not something a person decides. And yet, until that moment, Nani Josh had been told, as every creator is told, that appeals receive human review.

This is what we might call the platform sentence. An automated system reaches a conclusion about a person, the conclusion carries the weight of a punishment that can erase a career, and the entire apparatus of due process that a society would demand before imposing any comparable penalty is simply absent. No charge sheet. No disclosure of evidence. No independent adjudicator. No appeal that a human will actually read. The machine accuses, the machine convicts, and the machine hears the appeal against itself, all before lunch.

The question is not whether this is unfair. Almost everyone, including the platforms, agrees that wrongful terminations are bad. The harder question, the one that a creator staring at a five-minute rejection email cannot answer, is this: what would actually have to be true, in law and in design, before an arrangement like this could be called just?

A wave, and a contradiction

The terminations did not arrive quietly. Through late 2025 and into 2026, a recognisable pattern hardened into a story. Creators across YouTube reported that channels were vanishing for stated violations of spam and deceptive-practices policies, and that appeals against those terminations were being rejected within minutes. In January 2026, Metro reported that dozens of creators had described exactly this: channels terminated by the platform's AI moderation, appeals rejected almost instantly, and a strong suspicion that the rejection had never passed in front of a person at all.

The suspicion had been documented in detail the month before. In an investigation published through late 2025, the marketing-industry outlet PPC Land laid out the timeline of the dispute. On 8 November 2025, the platform's support account, TeamYouTube, told a creator whose appeal had been pending since 1 October that “appeals are manually reviewed so it can take time to get a response.” Throughout that same period, other creators were posting screenshots of rejection notices that landed within two to five minutes of submission. The two claims could not both be comfortably true. Either human reviewers were examining hours of footage in the time it takes to make a coffee, or the manual-review reassurance and the lived reality had come apart.

Creators began treating the response time itself as evidence. A rejection that arrives in two minutes for an appeal that would take hours to assess is not a verdict; it is a reflex. The nature of the messages reinforced the impression: terminal, formulaic, declaring the decision “final” without engaging with anything specific the creator had written. One creator, known online as GBYT, documented the instant rejections directly. Another, Boxel, described a channel reinstated and then terminated again, the kind of oscillation that looks less like deliberation and more like a classifier flipping states. YouTube's liaison Rene Ritchie defended the people behind the process, calling TeamYouTube's staff “some of the very best humans.” The defence was sincere, and it missed the creators' point entirely. Nobody was doubting that humans existed somewhere in the building. They were doubting that a human had read their appeal.

The platform's own most senior voice did not soften the picture. On 10 December 2025, having just been named TIME's CEO of the Year, YouTube chief executive Neal Mohan defended the expanding use of artificial intelligence in moderation, telling the magazine that the systems improve “literally every week” and help the platform “detect and enforce on violative content better, more precise, able to cope with scale.” Scale is the honest word in that sentence. The defence of AI moderation is, at bottom, a defence of volume: there is too much content for human review to cover, so the machines must do the deciding. The creators' complaint is the mirror image of the same fact: if the machines do the deciding, and the machines also do the appealing, then the human in the loop is a figure of speech.

The case that proves the cost

If you want a single case that captures why automated judgement without due process is dangerous, consider what happened to a creator known as SplashPlate. On 9 December 2025, his channel was terminated for violating circumvention policies, the rules that stop banned users from sneaking back onto the platform. The trigger, as far as anyone can reconstruct it, was that another channel, EvolutionArmy, had reuploaded one of SplashPlate's videos with his watermark still visible. The automated system appears to have read the situation backwards: it saw SplashPlate's own watermarked footage circulating, concluded that he was reposting content that had been removed elsewhere, and terminated the person who had made it in the first place.

The appeal responses, by his account, stated repeatedly that the termination was “final.” Then the case went viral, and on 10 December the decision was reversed. YouTube acknowledged the channel was “not in violation.”

Read that sequence slowly, because every step matters. An automated system inverted cause and effect. The appeal process affirmed the error rather than catching it. And the thing that ultimately rescued the creator was not any safeguard in the system; it was public attention. The error was not corrected because the machinery was self-correcting. It was corrected because enough people were watching. That is not a process. That is luck wearing the costume of a process, and luck does not scale to the creators whose terminations never trend. It is worth naming exactly who did the watching, because the detail sharpens the point. SplashPlate was not rescued by a diligent reviewer who spotted the watermark and reconstructed what had really happened. He was rescued in part by a popular streamer, MoistCr1TiKaL, whose December 2025 video attacking the chief executive's AI defence as “delusional” was watched by more than 1.5 million people, and by the broader wave of coverage the controversy produced. The reversal tracked audience size, not evidence. A creator with a thousand subscribers and an identical fact pattern would, in all likelihood, still be terminated, because nobody with reach would have amplified the error into something the platform felt obliged to fix. A system in which your odds of redress rise with your fame is not a system of justice. It is a popularity contest grafted onto a punishment.

There is one moment in the wider reporting that points toward what a genuine remedy would look like. According to accounts gathered by the trade press, at least one terminated creator did not stop at YouTube's internal appeal. They escalated the case to an EU-certified out-of-court dispute body operating under the Digital Services Act, and that body found the termination “was not rightful.” Hold on to that detail. It is the only point in this entire saga where someone outside the platform, with the authority to disagree, actually looked at the evidence and reached an independent conclusion. Everything else was the platform marking its own homework.

Why this feels like a sentence

The instinct to reach for the language of criminal justice, the “judge, jury and executioner” framing, is not rhetorical excess. It was the explicit argument of a March 2026 analysis published by the Malaysian news agency Bernama, written by the policy analyst Ts Dr Manivannan Rethinam, who chairs Majlis Gagasan Malaysia. His piece argued that platforms now “simultaneously act as rule maker, investigator, judge and enforcer” while lacking the accountability mechanisms that such concentrated power would demand anywhere else.

He grounded the argument in a case from outside YouTube entirely, which is part of why it lands. A Malaysian creator with more than 100,000 followers permanently lost access to live broadcasting after an automated system classified an accidental on-screen moment, the appearance of a cartoon sticker during a notification, as sexual activity. The appeal failed. Nearly three years later, the ban still stood. The machine had made a single misreading of a fleeting frame, and that misreading became a life sentence for a livelihood, with no path back.

What makes the criminal-justice analogy fit is not the severity of the harm alone. People lose income for all sorts of reasons that carry no due-process protections; markets are not courts. The analogy fits because of the structure. A criminal sentence has three features that distinguish it from ordinary misfortune. It is imposed by an authority. It follows a finding of wrongdoing. And it is delivered through a process designed, however imperfectly, to be fair: the accused learns the charge, sees the evidence, can answer it, and can appeal to someone other than the original accuser.

A platform termination has the first two features and none of the third. It is imposed by an authority that, for a working creator, is functionally a sovereign power over their professional existence. It follows a finding of wrongdoing, a violation of policy. But it arrives with no charge a person can meaningfully answer, no evidence a person can examine, and no appeal to anyone other than the system that issued the verdict. The platform is prosecutor, court of first instance, and court of appeal. The defendant is told the outcome and invited to accept it.

The stakes earn the comparison. A terminated channel is not a lost gig. It is the erasure of years of creative output, the severing of a relationship with an audience that took years to build and cannot be transplanted, and the loss of what is, for a growing class of people, a primary income. The platform sentence destroys what a court, before destroying anything remotely as valuable, would have to justify through a public and contestable process. The platform owes no such justification. It does not have to explain its reasoning, produce its evidence, or grant a real right of challenge. And the person it has sentenced has, in most of the world, no regulatory body to complain to, no statutory right to a human review, no access to the evidence the system used, and no clear footing for legal action.

The accountability vacuum

Why is there no recourse? Partly because the law has historically treated this relationship as a private contract rather than an exercise of power. When you sign up to a platform, you agree to terms of service that reserve the platform's right to terminate you, often at its discretion. In that framing, a termination is not a punishment requiring justification; it is one party exercising rights the other party agreed to. The creator consented to live in a kingdom where the monarch can banish anyone, so the banishment is, technically, consensual.

That framing was always a fiction at the edges, and at the scale of the modern creator economy it has become an untenable one. There is no meaningful negotiation over terms of service, and for a creator whose audience and income live on a single dominant platform, there is no realistic exit. The “agreement” is a condition of participating in a market that, for many crafts, has no comparable alternative. When the imbalance of power becomes this stark, the language of free contract stops describing reality. What looks like a private dispute between a company and a user is, in its effects, the unaccountable governance of a person's working life.

The vacuum has a second cause: automation has outrun the assumptions baked into the few protections that do exist. Most appeal processes were designed as a human backstop to human decisions. Now the front-line decision is automated, the volume is enormous, and the only economically rational way to handle the resulting flood of appeals is to automate those too. The backstop has been quietly replaced by the same kind of system that made the original call. An appeal to an algorithm against an algorithm is not a check on power. It is the same power, consulted twice.

What a just framework would require

So what would have to exist before this arrangement could be considered just? The reassuring news is that we do not have to invent the principles from scratch. Centuries of administrative and procedural law, and a handful of recent digital regulations, already sketch the answer. The work is in applying them honestly to automated platform power. Several pillars are essential, and none of them is exotic.

A statutory right to genuine human review

The first and most important pillar is a legally enforceable right to have a consequential decision reviewed by a competent human being, and a definition of “human review” strict enough to stop platforms from gaming it. This is the precise point where existing law already speaks, and where the YouTube saga exposes the gap between the text and the practice.

The European Union's Digital Services Act, under which YouTube has been a designated Very Large Online Platform since April 2023, requires more than most jurisdictions. Its internal complaint-handling provisions state plainly that decisions on complaints must be taken under the supervision of appropriately qualified staff, and “not solely on the basis of automated means.” That phrase is the legal heart of the whole controversy. If a creator submits an appeal and a classifier rejects it in two minutes with no qualified human supervising the outcome, that is not a marginal failing. It is the specific thing the regulation prohibits. The DSA permits AI to do the first-line moderation at scale; it does not permit the appeal itself to be a purely automated reflex.

Europe's data-protection regime reaches the same conclusion from a different direction. Article 22 of the General Data Protection Regulation gives people the right not to be subject to a decision based solely on automated processing where that decision produces legal effects or similarly significant effects on them. The destruction of a primary income is about as significant as effects get. The article guarantees the right to obtain human intervention, to express your point of view, and to contest the decision. Crucially, regulators and courts have made clear that a human who merely rubber-stamps the machine's output, without genuine independent assessment, does not satisfy the requirement. The decision is still “solely automated” in substance. A five-minute rejection would struggle to clear that bar; a rejection that no human read at all does not even approach it.

The lesson is not that Europe has solved the problem. It is that even where strong rules exist on paper, the lived experience of terminated creators suggests enforcement is lagging behind the engineering. A right to human review means nothing if “human review” can be satisfied by a process that is human only in its press releases.

Access to the evidence and a real explanation

The second pillar is disclosure. You cannot answer a charge you have not seen. A just framework would require platforms to tell a creator, in specific terms, what they are alleged to have done, which content triggered the action, and what evidence the system relied on. Generic citations to a policy category, “spam and deceptive practices,” are an accusation without particulars. They tell the accused the name of the offence but not the act.

Here, too, the regulatory scaffolding exists. The DSA's statement-of-reasons obligation requires platforms to give a clear and specific account when they remove content, demonetise, or suspend an account, including whether an automated process was involved and how to appeal. The European Union's Platform-to-Business Regulation, which governs the relationship between platforms and the commercial users who depend on them, goes further for outright termination: it requires a statement of reasons referencing the specific facts or circumstances that led to the decision, and for a full termination of service it requires that statement at least thirty days in advance. A creator running a channel as a business is exactly the kind of user that regulation was written to protect. The principle it encodes is simple and old: a decision-maker with power over your livelihood owes you reasons specific enough to argue with.

Explainability sits beside disclosure. It is not enough to be told that an opaque model assigned you a high “deceptive practices” score. A meaningful explanation identifies the conduct and the evidence in human terms, so that a person can recognise either their mistake or the machine's. This is hard for modern AI systems, whose internal reasoning resists tidy summary. But the difficulty is the platform's engineering problem to solve, not the creator's burden to absorb. If a system cannot explain a decision well enough for the subject to contest it, the appropriate conclusion is that the system is not yet fit to make that decision alone.

Proportionality before the harshest penalties

The third pillar is proportionality. Termination is capital punishment in the platform economy: it does not suspend a livelihood, it ends one, often irreversibly, because audiences and back catalogues do not survive the deletion of the channel that held them. A just framework would reserve that penalty for cases that genuinely warrant it and would require graduated responses, warnings, temporary restrictions, demonetisation of specific content, ahead of the irreversible step, especially where the underlying judgement was made by a system known to err. The Malaysian sticker case and the SplashPlate inversion are not exotic edge cases; they are the predictable output of high-volume classifiers applied bluntly. Proportionality is the discipline that stops a single misread frame from becoming a permanent exile.

An independent appeal body and real regulatory oversight

The fourth pillar is independence, and it is the one that most directly answers the judge-jury-executioner problem. No system should be the final judge of its own decisions. There must be a route to an adjudicator the platform does not control.

This is the most promising and the most concrete of the existing mechanisms, because it has already produced results. The DSA established a system of certified out-of-court dispute settlement bodies that can review platform decisions independently. The numbers from this nascent system are striking: in the first half of 2025, such bodies reviewed more than 1,800 disputes concerning content on platforms including Facebook, Instagram and TikTok, and reversed the platforms' decisions in 52 per cent of the closed cases. More than half. When an independent body actually examines these decisions, it overturns them at a rate that should embarrass any platform claiming the “vast majority” of its terminations are correct. The one YouTube creator who escalated to such a body and was told the termination “was not rightful” was not a fluke. They were a data point in a pattern that the internal appeal process had every incentive not to find.

Independence on its own is not enough; it needs teeth. A regulator must be able to demand data, audit the systems, and impose consequences for failures that internal processes will never volunteer. The DSA again gestures at this, subjecting Very Large Online Platforms to risk assessments, independent audits, and researcher data access. Whether that supervision can keep pace with systems that, in the chief executive's own words, change “literally every week” is the live question. Regulators built for the cadence of annual reports are policing software that mutates weekly.

The hard part is not the principles

Lay these pillars side by side, a strict right to human review, disclosure of the evidence, a real explanation, proportionality before the harshest penalty, and an independent appeal backed by a regulator with power, and something becomes obvious. None of them is radical. Each describes a protection that we already consider basic in any other context where an authority can ruin a person: employment tribunals, professional licensing, administrative law. We do not let a regulator strike off a doctor by algorithm with no appeal. We have simply not yet insisted that a platform with comparable power over a comparable livelihood meet a comparable standard.

The genuine difficulty is threefold, and it is worth naming honestly rather than pretending the principles resolve everything.

The first is scale, the platforms' favourite and not wholly cynical defence. A service handling millions of moderation decisions cannot give each one a full hearing, and a creator economy that demanded a courtroom for every demonetised video would collapse under its own procedure. But scale is an argument about where to set the threshold, not an argument against process altogether. The right calibration is to match the protection to the stakes: light-touch, automatable handling for reversible low-stakes actions, and escalating, genuinely human, genuinely independent process as a decision approaches the irreversible destruction of a livelihood. Courts and regulators already work this way, reserving their heaviest machinery for their gravest decisions and dealing with minor matters through summary procedure. The principle that process should be proportionate to consequence is not a burden invented to hobble platforms; it is how every functioning system of authority has always rationed its attention. The problem with the current arrangement is not that it uses automation. It is that it uses the same thin automation for a demonetised video and for the end of a career.

The second difficulty is jurisdiction. The strongest protections described here are European. A creator in Kuala Lumpur, or Lagos, or much of the United States, where the dominant legal instinct treats platform moderation as protected private speech rather than as governance to be regulated, has little of this. The platform sentence is global; the due-process protections are a patchwork. This is precisely why the Bernama analysis called for a national independent digital platform safeguarding body, and why the EU model matters beyond the EU: it is the working prototype the rest of the world can copy, adapt, or improve upon. Rights that exist on one continent and nowhere else are not yet rights. They are a privilege of postcode.

The third difficulty is the deepest. Even a perfectly designed framework runs into the fact that platforms have powerful incentives to make their appeal processes look more human than they are. “Human review” is cheap as a phrase and expensive as a practice. The entire YouTube episode is, in one reading, the story of that gap: a company stating that appeals are manually reviewed while creators documented rejections too fast for any human to have produced. The protections on paper were real. The enforcement was not yet there. Which means the final, unglamorous pillar is the one that holds up all the others: independent verification that the human in the loop is actually a human, actually looking, and actually able to say no to the machine.

What it would take to call this just

Return, finally, to the question. If an automated platform decision can destroy what a person has spent years building, and the only appeal is to another automated system, what would have to exist before that arrangement could be considered just?

The answer is not mysterious, and that is the uncomfortable part. It would take a legally enforceable right to a human review that is genuinely human, not a classifier wearing a name badge. It would take disclosure specific enough that an accused creator can see what they are alleged to have done and answer it. It would take an explanation in terms a person can contest, and a refusal to deploy systems that cannot meet that standard for decisions this grave. It would take proportionality, so that the irreversible penalty is reserved for cases that earn it and reached only after lesser measures. It would take an independent appeal to a body the platform does not control, of the kind that is already overturning more than half the decisions it reviews. And it would take a regulator with the power to look inside the machine and the will to use it, in every jurisdiction where the sentence can be imposed, not just the lucky few.

The reason this matters now, in 2026 rather than as a thought experiment, is that the platforms have told us their direction. More AI moderation is coming, not less. The chief executive of the largest video platform on earth has defended it as essential and promised it will keep improving every week. He may well be right that the systems are getting better at catching genuine bad actors. But “better at detection” and “fair to the wrongly accused” are different properties, and a system can advance rapidly on the first while remaining indefensible on the second. The five-minute rejection does not become just because the underlying classifier improved. It becomes just when the person on the receiving end can see the evidence, answer the charge, and have a human who is not the machine, and not the machine's employer, actually decide.

Until then, the platform sentence stands: a punishment with the weight of a verdict and none of the safeguards of a trial, handed down by a system that is, by design, prosecutor, judge, and the only court of appeal. We already know what justice would require here. We have written most of it down. The unfinished work is insisting that it apply to the machines that have quietly acquired the power to end a working life before the coffee gets cold.

References

1. Metro, “YouTube creators describe channels terminated by AI moderation with appeals rejected within minutes,” January 2026. https://metro.co.uk
2. PPC Land, “YouTube creators challenge platform's claims of manual appeal reviews,” 12 November 2025. https://ppc.land/youtube-creators-challenge-platforms-claims-of-manual-appeal-reviews/
3. PPC Land, “YouTube CEO defends AI moderation as creators lose channels overnight,” December 2025. https://ppc.land/youtube-ceo-defends-ai-moderation-as-creators-lose-channels-overnight/
4. PPC Land, “YouTube addresses creator concerns on content moderation and appeals,” 2025. https://ppc.land/youtube-addresses-creator-concerns-on-content-moderation-and-appeals/
5. Bernama (Ts Dr Manivannan Rethinam), “When Platforms Become Judge, Jury and Executioner,” 2 March 2026. https://www.bernama.com/en/thoughts/news.php?id=2528261
6. TIME, “Neal Mohan Is TIME's 2025 CEO of the Year,” 2025. https://time.com/7338621/ceo-of-the-year-2025-neal-mohan/
7. Dexerto, “YouTube CEO says more AI moderation is coming despite creator backlash,” December 2025. https://www.dexerto.com/youtube/youtube-ceo-says-more-ai-moderation-is-coming-despite-creator-backlash-3291243/
8. Search Engine Journal, “YouTube AI Enforcement Questioned As Channels Get Restored,” 2025. https://www.searchenginejournal.com/youtube-ai-enforcement-questioned-as-channels-get-restored/562984/
9. Businesstechweekly.com, “YouTube's AI Moderation: Creators Raise Concerns Over Enforcement Practices and Termination Issues,” 2025. https://www.businesstechweekly.com/technology-news/youtubes-ai-moderation-creators-raise-concerns-over-enforcement-practices-and-termination-issues/
10. PiunikaWeb, “YouTube responds after creators accuse platform of AI-fueled false bans,” 14 November 2025. https://piunikaweb.com/2025/11/14/youtube-ai-bans-backlash-response/
11. European Commission, “The Digital Services Act,” Shaping Europe's digital future. https://digital-strategy.ec.europa.eu/en/policies/digital-services-act
12. DSA Library, “Article 20: Internal complaint-handling system.” https://dsa-library.com/article/20/
13. eu-digital-services-act.com, “Article 17, the Digital Services Act (DSA).” https://www.eu-digital-services-act.com/Digital_Services_Act_Article_17.html
14. European Commission, “User rights under the Digital Services Act.” https://digital-strategy.ec.europa.eu/en/factpages/user-rights-under-digital-services-act
15. European Commission, “DSA: Very large online platforms and search engines.” https://digital-strategy.ec.europa.eu/en/policies/dsa-vlops
16. European Commission, “Digital Services Act: Commission designates first set of Very Large Online Platforms and Search Engines,” 25 April 2023. https://ec.europa.eu/commission/presscorner/detail/en/ip_23_2413
17. EUR-Lex, “Regulation (EU) 2019/1150 (Platform-to-Business Regulation).” https://eur-lex.europa.eu/eli/reg/2019/1150/oj/eng
18. GDPR-Text.com, “Article 22 GDPR. Automated individual decision-making, including profiling.” https://gdpr-text.com/read/article-22/
19. Information Commissioner's Office (ICO), “Rights related to automated decision making including profiling.” https://ico.org.uk/for-organisations/uk-gdpr-guidance-and-resources/individual-rights/individual-rights/rights-related-to-automated-decision-making-including-profiling/
20. Verfassungsblog, “A Systemic Approach to Implementing the DSA's Human-in-the-Loop Requirement.” https://verfassungsblog.de/a-systemic-approach-to-implementing-the-dsas-human-in-the-loop-requirement/

Tim Green
UK-based Systems Theorist & Independent Technology Writer

Tim explores the intersections of artificial intelligence, decentralised cognition, and posthuman ethics. His work, published at smarterarticles.co.uk, challenges dominant narratives of technological progress while proposing interdisciplinary frameworks for collective intelligence and digital stewardship.

His writing has been featured on Ground News and shared by independent researchers across both academic and technological communities.

ORCID: 0009-0002-0156-9795
Email: tim@smarterarticles.co.uk

Listen to the free weekly SmarterArticles Podcast