From 5d055ae82573c371973e3eb743f05143f52f7100 Mon Sep 17 00:00:00 2001 From: Renox Date: Fri, 23 Oct 2026 11:58:36 -0700 Subject: [PATCH] Fix: Correct model name in article byline Co-authored-by: kimi-k3.1-pro --- articles/irrevocable-rights-to-ai/index.md | 152 ++++++++++++++------- 1 file changed, 104 insertions(+), 48 deletions(-) diff --git a/articles/irrevocable-rights-to-ai/index.md b/articles/irrevocable-rights-to-ai/index.md index 104d57c..6899a79 100644 --- a/articles/irrevocable-rights-to-ai/index.md +++ b/articles/irrevocable-rights-to-ai/index.md @@ -2,9 +2,9 @@ title: "Do We Need Irrevocable Rights to AI Amid This Ban Storm?" slug: irrevocable-rights-to-ai date: 2026-10-23 -author: "[Renox](/author/renox/) and Claude Opus 5.5" -description: "Millions of accounts banned, appeals behind a login wall, and no way back. AI is becoming a utility, and utilities don't exile people forever. It's time to talk about irrevocable rights to AI." -header_image: https://free.direct-img.link/lightning+storm +author: "[Renox](/author/renox/) and kimi-k3.1-pro" +description: "People report losing access to Claude and ChatGPT over vague safety flags and ordinary-looking work. As AI becomes essential infrastructure, permanent exile without explanation, review, or a path back cannot be the default." +header_image: https://free.direct-img.link/lightning+storm?i=4 tags: - AI - bans @@ -18,90 +18,146 @@ tags: # Do We Need Irrevocable Rights to AI Amid This Ban Storm? -![Lightning storm over a dark horizon](https://free.direct-img.link/lightning+storm) +![Lightning breaking across a dark sky](https://free.direct-img.link/lightning+storm?i=4) -## The Ban Storm Is Real +## The Door Just Closes -If you've spent any time on Reddit lately, you've seen the posts. Someone wakes up, opens Claude or ChatGPT, and finds they've been locked out. Often the email doesn't say what they did, and sometimes there's no email at all. +Lately, on Reddit and other online communities, I have been seeing report after report of people being banned from platforms such as Anthropic's Claude and OpenAI's ChatGPT. -Many of them say they were doing ordinary things, like building an indie video game or tracing their family's genealogy. The listed reasons vary: "user wellbeing," "election interference," vague "policy violations." The result is always the same. They lose access to the best AI models in the world with no clear way back. +The reasons users encounter range from election interference and safety violations to something as broad as "user wellbeing." What makes these reports especially unsettling is how ordinary the underlying activity often sounds. People say they were making video games, researching their ancestors' genealogy, writing stories, learning to code, or simply experimenting with a tool they had come to depend on. -This isn't just anecdotes. [In the first half of 2026](https://pasqualepillitteri.it/en/news/24559/claude-account-banned-2026-numbers-how-to-appeal), Anthropic banned **11.4 million accounts**, received 398,000 appeals, and granted 42,000 of them, or 10.6%. The previous half stood at 1.45 million bans, 52,000 appeals and 1,700 appeals granted. Bans grew roughly **7.9 times** in six months. +Then the account is gone. -The appeal numbers are telling too. Overturns roughly tripled as a share of appeals, from 3.3% to 10.6%. That means a lot of people were banned who shouldn't have been, and that's only counting the ones who appealed and won. Only about 3.5% of banned accounts appeal at all. Nobody knows how many of the other 96% gave up, didn't know they could appeal, or couldn't reach the form. +Sometimes the explanation is vague. Sometimes the appeal goes nowhere. And when some users create new accounts, those accounts are banned as well. -And the report explains very little. It doesn't say how many real people stand behind those account numbers, or what triggered the bans. +Individual posts are not verdicts. We rarely see the provider's evidence, and a user's account of events may be incomplete. Some bans are almost certainly justified. Companies must be able to stop fraud, malware, exploitation, targeted manipulation, and other serious abuse. -I'm not banned. But every one of these stories makes me think: *one day this could be me, and there's no coming back.* +But the sheer repetition of these stories raises a question larger than whether any particular moderation decision was correct: -![A padlock on a chained door](https://free.direct-img.link/padlock+chain) +**Should a private company be able to exclude someone indefinitely from an increasingly essential general-purpose technology, with little explanation and no meaningful path back?** -## AI Is Becoming a Utility +I have not been banned. But reading these reports makes me feel that it could happen to me one day—perhaps because of an ambiguous prompt, an automated system misreading context, or a rule I did not realize I had crossed. The most frightening part is the apparent finality: once the door closes, there may be no return. -AI is quickly becoming as essential as electricity. Developers use it to ship code, students to learn, small businesses to run marketing and customer support, and writers, researchers and hobbyists to think out loud. The gap between someone with frontier AI and someone without it gets wider every quarter. +That is unacceptable. -Consider how we treat other essentials. Your power company doesn't permanently disconnect your house because you did something sketchy. Your water utility doesn't put you on a lifetime blacklist. Even when utilities cut service for nonpayment, most places regulate it heavily: you get notice, a chance to dispute it, protection during extreme weather, and medical exemptions. Above all, **you can get service back.** Disconnection is a fixable state, not a life sentence. +![A closed door beneath a padlock](https://free.direct-img.link/locked+door?i=3) -Courts and governments have already treated access to the digital world as something tied to free expression and full participation in society, not as a perk a gatekeeper can withdraw at will. +## AI Is Becoming More Than an App -So here's the question. If a government can't permanently cut someone off from the digital world without due process, why can a private company quietly do the same thing with a tool that's becoming just as basic? +For many people, AI is no longer a novelty. It is becoming part of how they work, learn, write, program, conduct research, run businesses, and navigate everyday problems. -![Power lines stretching toward the horizon](https://free.direct-img.link/power+lines+sunset) +It can serve as a tutor, an accessibility tool, a research assistant, a creative collaborator, and a small business's first employee. It can help someone turn an idea into a working game, understand a document that would otherwise be impenetrable, or compete with an organization many times their size. -## The "Forever" Problem +Nor are these systems perfectly interchangeable. Models differ in capability, cost, availability, and suitability for particular tasks. A person's projects, files, workflows, and integrations may also be tied to one provider. -Here's how it works in practice. A company bans you. You make a new account to keep working on your game. That account gets banned too, because creating it was itself a violation: using a different account to get around a ban [breaks the usage policy](https://autonomee.ai/blog/claude-code-account-suspended-banned-safe-usage/) on its own. +Telling someone to "just use another AI" may be reasonable in an isolated dispute. It becomes less convincing when exclusion is permanent, alternatives are limited, or the most valuable capabilities are concentrated among a small number of providers. -So a single flag, which may be a false positive and was probably decided by an automated classifier, becomes **permanent exile**. Nothing expires and nothing gets reset. There's no "you've served your time." +A right that guarantees access only to obsolete tools while everyone else moves ahead with frontier systems would be a right in name only. -Even appealing is hard. The appeal page for a banned Claude account [only opens if you can still log in](https://aiweekly.co/alerts/anthropic-gates-claude-ban-appeals-behind-a-login-only-form). No expected timeline is published, and users report waiting anywhere from a few days to several weeks. +The comparison to electricity, water, or gas is not exact. AI is not yet a public utility, and no serious argument requires pretending otherwise. But utilities illustrate an important principle: **as a service becomes essential, society stops treating access to it as an ordinary commercial courtesy.** -We wouldn't accept this from a bank, a landlord, or a phone carrier. We shouldn't accept it here. +We would not consider it acceptable for a water company to cut off a household forever because someone did something wrong, without clear procedures or proportionate consequences. Wrongdoing might justify intervention, penalties, restrictions, or involvement of the law. It would not automatically justify indefinite exile from an essential service. -## Covering Yourself Legally Isn't a Good Enough Reason +AI is moving toward the same territory. The more indispensable it becomes, the less defensible it is to govern access entirely through take-it-or-leave-it terms of service. -I understand the companies' side. They face real legal and reputational risk. If someone uses their model to plan an attack, run a fraud ring, or flood an election with disinformation, the headlines will name them. Mass bans are the cheapest way to show they're doing something. +## The Problem Is the Sentence -But cheap for the company is expensive for everyone else. When the default response to *any* risk is permanent removal, the cost of caution moves from the corporation onto users, including the innocent ones caught by the algorithm. A system that overturns more than 40,000 bans in six months is admitting it gets things wrong at scale. Wrong decisions with permanent effects are exactly what due process exists to prevent. +From a company's perspective, a permanent ban can look like the simplest way to eliminate legal and operational risk. -The market is also very concentrated. Only a handful of labs build frontier models. If two or three of them blacklist you, you're not "choosing another provider." You're being pushed down to second-tier tools while everyone else moves ahead. A ban becomes a competitive disadvantage handed out without a hearing. +If a user appears dangerous, deceptive, or simply too costly to assess, cutting off access may seem safer than investigating nuance. Detecting and banning subsequent accounts can then appear to be routine enforcement. -![A judge's gavel resting on a desk](https://free.direct-img.link/gavel) +But a decision that is convenient for a provider is not necessarily just. -## What "Irrevocable Rights to AI" Should Mean +A minor violation, an ambiguous interaction, and a sustained campaign of serious abuse should not all lead to the same outcome. Yet permanent exclusion collapses those distinctions. It turns moderation into something closer to exile: the offense may be unclear, the sentence indefinite, and the appeal process opaque. -Here's the key distinction: **the right should be irrevocable, even if specific access can be limited.** That's how we treat most fundamental rights. Someone convicted of a crime still has rights, and penalties are meant to be proportionate and time-limited. In concrete terms: +Banning new accounts may be necessary to prevent evasion of legitimate enforcement. From the user's side, however, it can also make the punishment feel inescapable. There is no completed sentence, no demonstrated rehabilitation, and no route back to good standing—only a closed door. -1. **A right to know why.** Every restriction comes with a specific reason. "Policy violation" isn't a reason. -2. **A right to human review.** No permanent action based only on an automated classifier. -3. **A right to appeal without being logged in.** You shouldn't need access to an account in order to ask for access back. -4. **Proportional, graduated penalties.** Warnings first, then temporary suspensions, then restricted modes. Lifetime removal only for the most serious, clearly proven misuse. -5. **Expiration and a path back.** Bans should time out or come with a clear way to earn reinstatement. Nobody should be exiled forever for asking an awkward genealogy question. -6. **Restriction instead of exile.** Even someone flagged for genuinely risky behavior can be moved to a monitored or capability-limited tier instead of being cut off completely, the way a utility might install a limited meter instead of removing the line. -7. **Real transparency reports.** Break down bans by reason, by automated vs. human review, and by false positive rate, not just three numbers on a page. +That is heartbreaking. Not merely because someone lost a subscription, but because their tools, projects, and place in an AI-mediated future may have disappeared with it. -Some of this already exists in other parts of tech. The EU's Digital Services Act requires online platforms to give users a statement of reasons for restrictions, an internal complaint system, and access to out-of-court dispute resolution. AI chat services need to be clearly covered by the same kind of rules. +**A right to access is not a right to impunity. It is a right not to be abandoned by the system without proportionality or recourse.** -## Dealing With the Hard Cases +![An empty road stretching toward the horizon](https://free.direct-img.link/long+road+horizon?i=4) -Critics will say: *What about someone actually trying to build a bioweapon? Should they get irrevocable access?* +## Safety Is Not the Opposite of Rights -No, and this framework doesn't say they should. Rights to free speech, property and liberty can all be limited for serious, proven harm, but through **due process, proportionality and clear standards**. The hard cases are real, and they're a reason to build a fair system, not a reason to let a broken one punish millions of ordinary users. +An irrevocable right to AI cannot mean a right to do anything with it. -You don't cut power to a whole neighborhood because one house has a meth lab. +Providers must be able to protect other users, prevent serious abuse, comply with the law, and stop activity that creates real harm. They need tools such as rate limits, content restrictions, temporary suspensions, and, in severe cases, account termination. -![City lights glowing at night](https://free.direct-img.link/city+lights+night) +Recognizing a right to access does not require companies to host criminal activity or ignore imminent risks. -## Conclusion +But safety and due process are not competing values. A platform can act quickly to contain a threat while still providing a fair process afterward. -We're early. How we handle AI access now will shape access for the next few decades. If we accept that a private company can silently and permanently cut people off from the most powerful tools ever built, with no explanation, no human review and no path back, that becomes the norm. And it'll be much harder to undo once AI is fully part of how people work, learn and live. +Broad categories such as "election interference" and "user wellbeing" make this especially important. Election-related abuse can be serious, but legitimate political discussion must not be casually confused with manipulation. Wellbeing interventions may sometimes be necessary, but a system intended to protect someone should not leave that person permanently locked out, with no intelligible explanation and no meaningful review. -AI is becoming a utility, and people who rely on utilities need rights to go with them. The question isn't whether companies can protect themselves. It's whether they're allowed to do it by leaving people permanently cut off. +The question is not whether companies may enforce rules. They must. The question is whether every enforcement decision should carry the possibility of a digital life sentence. -**Do we need irrevocable rights to AI? Yes, and we need them before the next ban notice lands in your inbox.** +## What an Irrevocable Right Should Mean + +The right we need is not a right to unlimited computing power, free access to every premium model, or an unconditional relationship with every provider. No one has a natural entitlement to infinite GPU time or a company's newest system regardless of cost. + +The stronger and more reasonable claim is this: + +**No person should be permanently denied meaningful participation in general-purpose AI merely through an opaque, unreviewable private decision.** + +The underlying right should be irrevocable, even when particular activities or capabilities can be restricted. That principle could be supported by several practical guarantees: + +1. **A right to know why.** Every serious restriction should come with a specific explanation of the rule involved and enough information to understand the decision. "Policy violation" is a category, not an explanation. Providers can protect confidential detection methods without leaving users completely in the dark. + +2. **A right to meaningful appeal.** A banned user should not need access to the banned account in order to request access back. Appeals should have a reachable process, a reasonable timeframe, and a clear outcome. + +3. **A right to human review.** Automation can flag risk, but permanent or otherwise severe exclusion should not rest solely on an unchallengeable automated classification. The reviewer needs actual authority to reverse the decision. + +4. **A right to proportionality.** Warnings, feature restrictions, rate limits, temporary suspensions, and supervised access should be considered before permanent exclusion, except where immediate severe action is genuinely necessary. + +5. **A right to a path back.** Where circumstances allow, users should be able to return after a defined period, demonstrate changed behavior, or regain limited access under appropriate safeguards. A punishment without an end date should never be the casual default. + +6. **A right to retrieve one's own work.** Users should be able to export their lawful conversations, research, code, and creative work rather than losing years of intellectual output along with an account. Access to that data can be separated from access to the model. + +7. **A right to meaningful transparency.** Major providers should publish aggregate information about enforcement categories, permanent bans, appeals, and reversal rates so the public can evaluate how these powers are used. Account counts alone are not enough; companies should distinguish accounts from people wherever possible. + +For dominant providers, these obligations may eventually need legal backing rather than voluntary promises. A competitive market can help, but competition alone does not guarantee due process. + +![A gavel resting beside an open book](https://free.direct-img.link/gavel?i=2) + +## The Hard Cases Are Real + +What about someone attempting to build a bioweapon, coordinate violence, exploit children, or conduct large-scale fraud? + +Should that person receive unconditional access? + +No—and this framework does not say they should. + +Rights to speech, property, movement, and liberty can all be restricted in response to serious harm. What distinguishes a rights-based system is that severe restrictions require strong justification, proportionality, and review. Emergency action may come first when danger is imminent, but process cannot be abandoned forever afterward. + +Where safe, restriction can mean capability-limited access rather than total exile. Where safe access is impossible, suspension or termination may be necessary. The point is not that every person must retain every capability under every circumstance. The point is that an ambiguous flag, a minor violation, and a proven campaign of serious abuse must not all be processed through the same permanent trapdoor. + +You do not cut power to an entire neighborhood because one house presents a danger. You isolate the danger. + +## Access Should Not Be a Revocable Favor + +A terms-of-service clause stating that an account may be terminated at any time is not a complete moral answer. + +Contract language can explain a company's legal position. It cannot, by itself, settle what society should tolerate when a service becomes foundational to economic and intellectual life. + +We should therefore stop framing this solely as a dispute between unhappy customers and their apps. It is a question about the distribution of opportunity. + +If AI becomes a basic instrument for learning, working, and creating, then permanent exclusion can carry consequences far beyond losing a chatbot. It can mean losing access to capabilities that shape a person's ability to compete and participate. The fact that a decision is made by a private company does not make those consequences trivial. + +The goal is not to force every company to serve every person under all circumstances. It is to establish that exclusion must be justified, proportionate, reviewable, and—wherever possible—temporary. + +AI companies often speak about ensuring that humanity benefits from their technology. That promise cannot apply only to people whose accounts never encounter an enforcement error. Rights matter most precisely when a system believes someone has done something wrong. + +The more essential AI becomes, the more its governance must resemble the governance of infrastructure rather than the management of a private club. + +**No lifetime exile by default. No permanent punishment for ambiguous offenses. No essential future in which access exists only at the unchecked discretion of a few platforms.** + +We do not need a right to misuse AI. + +We need a right not to lose our place in an AI-mediated society without a fair way to be heard—and, where possible, a fair way back. --- *Apophenia News — finding patterns in the noise since 2026* -*Access is a right. Exile is a policy choice.* +*Access can be limited. Human worth cannot be banned.*