Polski
Content & AI search · Claude

Claude from 12 November: suspicion is enough. What the new rules change

Anthropic is changing its Usage Policy on 12 November 2026. The terms of service stay as they are, but the policy gets a new suspicion threshold for bans, a longer list of platform abuses and a new section on deceptive campaigns. I take it apart piece by piece and check what applies to work in SEO and marketing.

Sebastian Baryś
Sebastian Baryś
18 October 2026 19 min read

Introduction

English version of an article first published in Polish on 18 October 2026. Read the Polish original.

On 12 November 2026 a new version of Claude’s Usage Policy takes effect. Anthropic published it on 8 October and writes in the announcement that most of the changes only clarify existing rules. Having compared it with the version of 15 September 2025, I see more changes than the announcement suggests, and some of them apply directly to people who work in SEO and marketing.

I will start with a distinction that is easy to get wrong. On 12 November the acceptable use policy and the regions policy change. The terms of service stay as they are. As of 18 October, the Consumer Terms (Claude.ai, Pro, Max) and the Commercial Terms (API) carry the same dates as before. The difference matters, because the policy says what you are allowed to do, while the terms say when and on what conditions Anthropic can close an account.

I read both versions of the policy, both sets of terms and the official pages on account bans. Below I show what changes, how it reads in the original, what follows from it in everyday work and what Anthropic does not explain. The screenshots come from the policy page, so you can check every quote yourself.

What you will learn:

  • Which documents change on 12 November and which stay as they are, and why they are easy to confuse.
  • How the wording on bans has changed: “if we learn” has given way to “if we suspect”, and warnings and access limits have been added.
  • What the new list of platform abuses means for multiple accounts, getting around the log-in process and tools that run on a subscription.
  • What the new section on deceptive campaigns is and why it touches site networks, fake reviews and hidden sponsorship.
  • Who must keep a human “in the loop” for content on health, law and finance, and who is not affected.
  • How to appeal a ban, what the statistics say and what Anthropic does not explain.
  • What I would do before 12 November, and in what order.

What exactly takes effect on 12 November?

On Anthropic’s website two documents carry the note “Effective November 12, 2026”: the Usage Policy and the Supported Regions Policy. The Consumer Terms still say “Effective October 8, 2025”, and the Commercial Terms “Effective June 17, 2025”. I found no announcement of a newer version of the terms in those documents.

What changes and what stays

changing

Usage Policy

effective 12 November 2026, previous version from 15 September 2025

changing

Supported Regions Policy

effective 12 November 2026, enforcement clarified

unchanged

Consumer Terms (Claude.ai, Pro, Max)

effective 8 October 2025

unchanged

Commercial Terms (API)

effective 17 June 2025

Position as of 18 October 2026, based on the dates shown on Anthropic’s pages.

The distinction matters because both sets of terms refer to the policy. The Consumer Terms say, in the section on use, that you may use the services only in compliance with the terms, including the Acceptable Use Policy and the regions policy. The Commercial Terms do the same by incorporating the policies into the agreement. So a change to the policy changes what you are allowed to do without changing a single sentence of the terms.

The Consumer Terms also provide that Anthropic gives at least 30 days’ notice of changes that may affect your use or your rights. A user who does not want to accept them can terminate the agreement before they take effect and receives a pro-rata refund for the unused period of a subscription bought on Anthropic’s website. The terms do not say explicitly whether this rule covers changes to the policy itself. The gap between the announcement and the effective date is 35 days, so it is consistent with that rule.

From the previous version to the new one

  1. 15 September 2025

    Previous Usage Policy

    the version I compare the new one with

  2. 8 October 2025

    Consumer Terms

    the current version of the consumer terms, with no newer one as of 18 October 2026

  3. 8 October 2026

    New policy announced

    Anthropic publishes the text and the post “2026 Usage Policy update”

  4. 12 November 2026

    New policy takes effect

    35 days after the announcement

The Usage Policy page on anthropic.com with the date Effective November 12, 2026, a link to the previous version and the first paragraph on who the policy applies to
The top of the Usage Policy page with the effective date. Screenshot from 18 October 2026.

How is the policy built and who does it cover?

The policy is divided into three layers. The layout is the same as in the previous version. The content of all three has changed, and the first has gained a section.

Universal Usage Standards

Prohibitions for everyone and for every use. Now 15 sections, previously 14.

High-risk Use Case Requirements

Requirements for uses that can affect health, rights, money or access to services. Rebuilt.

Additional Use Case Guidelines

Chatbots and agents that talk to people, products for minors, agentic uses, software from directories.

The scope is broad. The policy applies to anyone who submits inputs to Anthropic’s products. Anthropic lists four groups: people who use the apps, including Claude.ai and Claude Code, developers and companies using the API, customers who use Claude through cloud providers and authorised resellers, and end users of products and services that have Claude inside. So if you build a chatbot for a client on the API, its users are covered by the policy too.

One sentence from the start of the document strikes me as important for agencies and freelancers: the policy focuses on how you use the products, not on the industry you work in. Its prohibitions and requirements “do not categorically prohibit” use by any industry, provided the use itself complies with the policy. So finance, medicine and law are not excluded, but they come with additional requirements, which I cover below.

Account bans: what changes in enforcement?

I will start with account bans. I have put side by side two paragraphs that sit at the start of the document in both versions. The difference is a few words, but each of them changes something.

Trigger

Before: if Anthropic learns that you have violated the policy (learn)

After: if Anthropic suspects that you may have violated it (suspect)

Measures

Before: throttling, suspension, termination

After: warning, throttling, limiting access, suspension, termination

Blocked responses

Before: Anthropic may block or modify responses when a prompt violates the policy

After: encountering a real-time block “does not by itself indicate a violation”

Access with modified safeguards

Before: not mentioned in the policy

After: verification programmes: Cyber Verification Program and Life Sciences Verification Program

Excerpt from the new Usage Policy: if we suspect a violation, we may warn you or throttle, limit, suspend or terminate your access, plus the paragraph on real-time safeguards, where a block does not indicate a violation
The enforcement provision in the new version of the policy. Screenshot from 18 October 2026.

The threshold. Previously Anthropic could act when it learned of a violation. Now a suspicion that you may have committed one is enough. This is a change in the wording of the document, not proof of a change in practice. The text does not tell you whether there will be more bans. What it does tell you is that the document no longer requires a fact to be established before action is taken.

The ladder of measures. Warnings and access limits have joined the earlier measures. The policy does not say when each one applies or whether Anthropic has to start with the lightest. The Help Center adds that warnings may appear when prompts look as if they violate the policy, and that for the API they are tied to violation thresholds across the whole API account.

Measures listed in the policy

  1. 01 Warning new in the policy text
  2. 02 Throttling throttle, was there and still is
  3. 03 Limiting access limit, new in the policy text
  4. 04 Suspension was there and still is
  5. 05 Account termination was there and still is
The order is mine, starting with the mildest. The policy lists the measures without any commitment to apply them in sequence.

A blocked response is not an accusation. The policy now says that real-time safeguards may block or limit responses, but that encountering a block does not by itself indicate a violation. If the model refuses to answer, that does not mean you have broken the policy. For dual-use or high-risk applications in selected fields, Anthropic describes verification programmes through which you can apply for access with modified safeguards: the Cyber Verification Program and the Life Sciences Verification Program.

Then there are the terms, which work together with the policy. The Consumer Terms allow Anthropic to suspend or terminate access without notice when it believes there has been a material breach. With a subscription this means no refund, and Anthropic reserves the right to take further legal action and to claim compensation for losses. After terminating an account it may also delete the data linked to it. The combination of “we suspect” in the policy and “we believe” in the terms gives Anthropic wide latitude. That is not an accusation, just a description of an arrangement you need to know.

Platform abuse: the list gets longer

The “Do Not Abuse Our Platform” section is the one most often connected with bans. I compared it point by point with the old version. Two points are entirely new, two have been extended, one clarified and one narrowed.

Multiple accounts

extended

Added: a ban on creating multiple accounts to avoid detection, to get around product guardrails or to generate similar inputs that violate the policy. Simply having two accounts is not prohibited. Getting around a ban with a new account or someone else’s was already prohibited.

Sign-up and log-in

extended

Before: bypassing safeguards in order to make the model produce harmful outputs. Now: intentionally bypassing or interfering with product capabilities, restrictions and guardrails, including by getting around the sign-up or log-in process.

Payments

new

A ban on using stolen, fraudulent or unauthorised payment methods and on evading payment for the services.

Reselling and proxying

new

A ban on reselling, proxying or otherwise providing access to Claude through unauthorised means, including services that route requests through consumer subscriptions or misrepresent the product or client being used.

Regions

clarified

The ban on providing access in breach of the regions policy now names end users explicitly.

Training on outputs

narrowed

The ban now covers model outputs. Previously it covered inputs and outputs. The change narrows the ban.

The Do Not Abuse Our Platform section in the new Usage Policy: eight points, from coordinating activity across multiple accounts to reselling access through consumer subscriptions
The platform abuse section in the version effective from 12 November. Screenshot from 18 October 2026.

For freelancers and agencies the most important point is the one on reselling and proxying. The policy bans services that route requests through consumer subscriptions or misrepresent the product or client being used. The Consumer Terms point the same way: they ban access through bots and scripts unless you use an API key or Anthropic explicitly permits it, and they ban reselling the services. The practical conclusion is simple: if something is meant to run automatically, it should connect through the API, not through an ordinary account.

During 2026 third-party tools’ access to Claude subscriptions changed several times, and the accounts I found contradict each other. What is certain is the provision in the new policy, so I am not giving any list of tools here. Check the tool’s documentation and Anthropic’s official announcements.

Deceptive campaigns: the section for SEO and GEO

In my view this is the most interesting new element for the SEO industry, because it touches practices we have known for years. Anthropic has gathered into one section rules that used to be scattered across the chapters on elections, fraud, privacy and disinformation. In the announcement it writes that the section applies to “deceptive activity of any kind”, political and commercial.

The Do Not Engage in Deceptive Campaigns or Artificial Activity section: fake personas and media outlets, hidden sponsorship, networks of ostensibly independent sites, manipulating the sources used by search engines and AI systems, and tools for such campaigns
The new section on deceptive campaigns. Screenshot from 18 October 2026.

Below I have matched five provisions with their counterparts in SEO and marketing work as I know it from the Polish market. The second part of each card is my interpretation. The policy gives no industry examples and leaves the line in borderline cases unresolved.

What the policy says

Fake personas, accounts, media outlets and organisations, impersonation, sockpuppets, astroturfing, fake reviews, bots that claim to be human

Where you see it in SEO

Reviews on a Google Business Profile and on review sites that pretend to come from customers, comments and forum posts “from ordinary users”, outreach campaigns with invented senders

What the policy says

Concealing the sponsorship of content intended to influence public opinion or decision-makers

Where you see it in SEO

Articles and mentions that are paid for but presented as independent. The text leaves the line for an ordinary advertorial unresolved

What the policy says

Distributing content through networks of “independent” websites or accounts to hide a common source

Where you see it in SEO

Networks of satellite blogs and sites that pretend to be independent, including link-building sets run from one source

What the policy says

Manipulating the sources from which search engines and AI systems draw answers, by seeding them with content that misrepresents its origin, authorship or independence

Where you see it in SEO

Networks of sites corroborating the same claims so that an AI model treats them as independent sources. This practice is named here outright

What the policy says

Tools, systems and infrastructure for such campaigns, and selling them as a service

Where you see it in SEO

Building and offering clients kits for mass account creation, fake reviews or site networks

The fourth card deserves a paragraph of its own. You rarely see a model provider’s policy name so directly a practice I would call “black-hat GEO”: seeding the sources from which search engines and AI systems draw answers with content that misrepresents its origin, authorship or independence. The example in brackets is specific: networks of sites posing as independent sources that corroborate the same claims.

Two caveats, so as not to go beyond what the policy says. First, it does not follow that Anthropic reports your sites to Google or that a site network will be penalised in search results. The policy is about using Claude for such activities. Penalties in Google are a separate system. I write about how Google assesses third-party content placed on sites with a reputation in my article on parasite SEO in the EU, and about how Google describes the effort, originality and accuracy of content in my article on the new definition of main content.

Second, the policy does not prohibit using AI to create content. The line is hiding who is behind the content and how its credibility was manufactured: fake reviews, fake authors, hidden sponsorship, networks of “independent” sites. Your own article under your own brand, written with a model’s help and signed by a real author, is not touched by these provisions. The section on fraud does, however, contain a line about plagiarising or submitting AI-assisted work “without proper permission or attribution”. It was in the previous version and it is in the new one, and the policy does not elaborate on it, so I will not attempt to interpret it. I described how Anthropic marks generated text in my article on the watermark in Claude.

What about content on health, law and finance?

Anthropic writes in the announcement that the requirements for high-risk use cases have not changed, and that the section was rewritten to answer directly the question of who they apply to. Comparing the texts, I see several differences that are more than cosmetic. I present them without judging whether they “change the requirements”, because that depends on the definition.

Eleven high-risk areas

LegalMedicalFinancialCreditInsuranceHousingEmploymentEducation and credentialsAccess to healthcarePublic benefitsLegal status
The requirements apply to recommendations in these areas. Previously the policy listed seven categories.

Scope

Before: specific consumer-facing use cases. Now: anyone who uses the products to make recommendations in one of the high-risk areas, or to control hardware that performs physical actions.

Areas

Before: seven categories, including “media or professional journalistic content”. Now: eleven areas, with no media category, but with credit, public benefits, access to healthcare and legal status.

Who reviews

There is a definition of a “qualified person”: someone with the education or experience and, where the law requires it, a licence. They must have the authority to change the recommendation before it reaches the recipient.

Disclosure

Before: a notice that AI is being used, at least at the start of each session. Now: the person receiving the advice or subject to the decision must be clearly told that AI was used. There is no need to name Anthropic or Claude.

Exceptions

A new list of what the requirements do not cover: general and educational content, wellness, explaining existing advice, internal drafts and research, applying a fixed rule without judgement, and decisions that are fully in the person’s favour.

What does this mean in everyday work? For a writer or an agency it is the distinction between general content and advice for a specific person. An article on how a mortgage works or what solar panels involve is general content, and the policy explicitly excludes it as long as it does not apply the information to a particular person’s situation. A tool that suggests treatment based on a specific patient’s test results, or a bot that assesses a credit application, is a high-risk use case. That requires a human with the authority to change the recommendation, and a notice to the person concerned that AI was used.

The disappearance of the “media” category is worth noting. In the previous version it required a human and disclosure when content was generated automatically and published externally. In the new one it is not on the list. That does not mean you can publish without checking: the sections on disinformation and deceptive campaigns still apply, and the policy does not relieve you of responsibility for what you put out.

Agents, chatbots and the browser

The third layer of the policy has gained two provisions that apply directly to client work. The first: chatbots and AI agents that talk directly to external users must tell the person that they are dealing with AI. Until now it was enough to do this at the start of the session; now you can also do it in the product interface. If you build a chatbot for a client’s website, that notice is an obligation, not decoration.

The second provision concerns agents. The policy says that the user is responsible for ensuring that the agents they build or deploy, including the actions taken through tools, browsers and connected systems, comply with the policy. For anyone who gives Claude tasks in the browser or lets it work in client accounts (Google Ads, Search Console, CMS dashboards), this is a clear signal: responsibility for what the agent clicks, sends or publishes lies with the user.

Two things from other sections matter here in practice. The first concerns privacy: the policy bans aggregating personal data from multiple sources to build profiles for sale or sharing without people’s knowledge, where this serves prohibited purposes. So lead data enrichment as such is not named as prohibited, but compiling data about people without their knowledge for a purpose the policy rules out is. The second concerns outreach campaigns: accounts or messages that pretend to be people who do not exist fall within the section on deceptive campaigns.

The new policy also adds requirements for hardware controlled by the model: a qualified operator must be able to observe the device and stop it, and safety limits such as speed or force must be enforced independently of the model. For most readers of this blog that is a curiosity, but it shows the direction the policy is heading in: the more autonomous the agent, the more requirements for the human who runs it.

Can you be banned for abusing the model?

The section on cruel and harmful behaviour has gained a point that sounds the most unusual: a ban on “sustained and needless abusive or cruel behavior toward our models”.

What Anthropic says. The provision is meant for extreme cases in which a user repeatedly treats the model cruelly with no apparent purpose. It does not cover typical frustration, pushback, dark themes in fiction, or model testing and research. The main enforcement mechanism is meant to be Claude’s ability to end a conversation with a persistently abusive user, which is already possible in Claude.ai and Claude Code.

What Anthropic does not say. It does not define what “sustained” and “needless” mean, or whether anyone can receive a warning or a ban for such behaviour. The general enforcement measures at the start of the policy cover all its sections, so formally this one too. Anthropic stresses, however, that the emphasis is on ending the conversation. Ordinary irritation at a model that gets things wrong is, according to Anthropic, not what this is about.

What happens after an account ban?

The policy describes when Anthropic can ban an account, but not what happens next. You have to piece that together from the Help Center, and it changes faster than the policy.

According to the Help Center article, accounts are banned for repeated violations of the Usage Policy, for creating an account from an unsupported location and for breaches of the terms. If you think the ban is a mistake, you log in to the banned account at claude.ai and fill in the appeal form, which is available only after logging in. The Safeguards team investigates the case. The page says response times are “longer than normal” but gives no time frame. Free, Pro and Max users banned for violating the policy can export their data or delete their account from that screen without contacting Anthropic. The scope of the export may depend on the type of violation.

Bans and appeals, January to June 2026

11.4m

banned accounts

398k

appeals, or about 3.5% of the number of bans

42k

appeals upheld, or about 10.6% of appeals

Data from Anthropic’s Transparency Hub, page updated on 23 July 2026. I calculated the percentages myself. Anthropic does not define on that page what it counts as a “ban”, and appeals may relate to bans from earlier periods, so this is an order of magnitude, not a success rate.

These figures cover the period before the new policy, so they say nothing about how the scale of bans will change after 12 November. What they do show is the scale of the phenomenon: bans are not an exception, and appeals are sometimes upheld, though in a small share of cases. I found no response time for an appeal in the documents, nor any commitment to give reasons for an account ban.

Regions: who cannot use Claude?

The Supported Regions Policy also takes effect on 12 November. According to the announcement it introduces no new categories, because Anthropic added the restrictions on companies majority-owned by entities from unsupported countries a year earlier. The change is meant to clarify enforcement. The ban covers: people who are physically located in an unsupported region, including users and employees of entities from supported countries; entities incorporated or headquartered in such a region; and entities controlled, directly or indirectly, by persons or entities from such regions.

Poland is on the list of supported countries. Russia, Belarus and China are not on it. The practical consequences for Polish companies and agencies: a team member working remotely from a country off the list cannot use Claude while they are there, and the owners and ownership structure of the client you are building a solution for matter. Anthropic does not state an ownership threshold, so in unclear cases the decision has to rest on a lawyer’s opinion, not on my article.

What does the policy not say?

A list of things that remain open for me after reading. Each of them may become clear in practice or in Anthropic’s later announcements, so treat them as the position today.

Questions the documents leave unanswered

  • Whether the change from “learn” to “suspect” will change how bans work in practice, and how often warnings are used before it comes to suspension.
  • What the response time for an appeal is, and whether Anthropic gives reasons for an account ban.
  • What “sustained and needless” means in the ban on abusing the model, and whether anyone can be penalised for it beyond having the conversation ended.
  • How Anthropic decides borderline cases of sponsored content that is not a political campaign.
  • What the “Model Hardware Standard” referred to in the hardware section is, and what ownership threshold determines “control” in the regions policy.
  • Whether existing deployments, such as client chatbots, have to change anything before 12 November, and what exactly the disclosure obligations look like in practice.
  • Which third-party tools can use a Claude subscription today. The sources I found contradict each other.

What would I do before 12 November?

Below is the order I would start in. This is my recommendation as someone who works with AI tools in SEO every day, not a requirement of the policy.

01. Establish who logs in to Claude and how

One account per person, and a team plan or the API for a team. The Consumer Terms say you are responsible for all activity on your account, and the policy bans giving access to people who have been banned. This is my recommendation; there is no provision on sharing a login.

02. Move automations to an API key

The Consumer Terms ban access through bots and scripts unless it goes through an API key or has explicit permission. Scripts that log in to an ordinary account are the wrong route.

03. Check the tools that use your Claude account

If an app logged in to your subscription instead of the API, check its current documentation with the provider. The policy bans services that route requests through consumer subscriptions.

04. Review your content processes against the campaigns section

Reviews, comments, outreach, sponsorship, site networks. The control question: does someone who sees the content know who is behind it?

05. Separate general content from advice for a specific person

A guide to how credit works is excluded from the requirements. A chat that suggests a drug dose or a credit decision to a specific person is not. That needs a human and a notice that AI was used.

06. Add an AI notice to your clients’ chatbots

A chatbot that talks directly to external users must say that it is AI, at the start of the session or in the interface.

07. Before you need it: note down the appeal route and export what matters

The form is available only after logging in to the banned account. Exporting your account data from time to time is worth doing anyway, because the terms allow data to be deleted after an account is terminated.

I describe how I organise work with models in SEO in my article on a ChatGPT and Claude workflow for SEO.

My view

I read this policy with one question in mind: will anything change for an ordinary user? For most people, no. Anyone who uses Claude for work, writes with it and analyses data will usually not run into any of the new bans. The changes mainly affect people doing things they should not be doing anyway: networks of fake accounts, reselling access, campaigns posing as independent sources.

Two things strike me as important. The first is the phrase “if we suspect”. In itself it is reasonable, because at a scale of millions of accounts (11.4 million bans in six months) nobody establishes the facts before every decision. The problem is that nothing stands next to it to balance it: a response time for appeals, a commitment to give reasons for a ban, a clear procedure in the policy itself. It is good that Anthropic publishes statistics. The fact that appeals are upheld in about one case in ten may mean that the system works well, or that it is mainly people who really were right who appeal. I cannot tell the two apart from the data.

The second is the section on deceptive campaigns. For the SEO industry it is a signal that practices people describe as “grey” are starting to be written into model providers’ policies as prohibited. Anthropic justifies this with its own observations: networks of fake accounts and news sites run with Claude’s help. Anyone who has built a GEO strategy on networks of “independent” sites should read that point twice.

I know this industry, though, and I know how this usually ends: SEO specialists will quickly find a way. After all, nobody “spams” today, sorry, scratch that: today people “create valuable content at scale”. The industry line is that this is exactly what gets you cited. Is that true? You will have to answer that for yourselves :) For me, today, the answer is trivially simple. Anthropic’s policy does not ban writing a lot. It bans spam and hiding who is behind the content, and it does not say where exactly the line runs between high-volume production of valuable articles and spam. That is a judgement in a specific case, not a rule from a table. I recognise the old race in this: model providers and search engines put up a fence, and the industry checks where it is lowest.

This is not a policy against working with AI. It is a policy against hiding who is behind the content. As long as you are not hiding who is behind the content, the campaigns section does not apply to you. But I still do not know what an appeal will look like if someone decides that it does.

FAQ: common questions

What changes on 12 November 2026?

A new version of the Usage Policy (the acceptable use policy) and the Supported Regions Policy takes effect. Anthropic published them on 8 October 2026. The changes include a new section on deceptive campaigns, a longer list of platform abuses, rebuilt requirements for high-risk use cases, more precise provisions on weapons and surveillance, and a new ban on sustained, needless abuse of models. As of 18 October 2026, the terms of service (Consumer Terms and Commercial Terms) carry the same dates as before.

Are Claude’s terms of service changing?

As of 18 October 2026, no. The Consumer Terms for Claude.ai, Pro and Max are dated 8 October 2025, and the Commercial Terms for the API 17 June 2025. Both documents refer to the Usage Policy and the Supported Regions Policy, though, so a change to the policy changes what you are allowed to do under the terms. The clauses on suspending and terminating an account remain as they were.

Can you be banned for swearing at Claude?

Probably not, although the policy does not rule it out explicitly. The ban covers “sustained and needless” abusive or cruel behaviour towards models. Anthropic writes that it is meant for extreme, repeated cases with no apparent purpose, not for ordinary frustration, pushback, dark themes in fiction or testing the model. The main enforcement mechanism is meant to be Claude ending the conversation on Claude.ai and in Claude Code. In general the policy allows warnings, limits, suspension and termination of an account, but it does not define a threshold for this section.

Can you have two Claude accounts?

The policy does not ban simply having two accounts. It bans creating multiple accounts to avoid detection, to get around product guardrails or to generate similar inputs that violate the policy, as well as getting around a ban with a new account or someone else’s. This is the position as of 18 October 2026. It is safer to keep one account per person and use a team plan or the API for a team, but that is my recommendation, not a provision of the document.

Can I use my Claude subscription in third-party tools?

It depends on the tool and on how it logs in. The new policy bans reselling and proxying access to Claude “through unauthorized means”, including services that route requests through consumer subscriptions or misrepresent the product or client being used. The Consumer Terms also ban access through bots and scripts unless you use an API key or Anthropic explicitly permits it. The status of specific tools changed during 2026 and secondary sources contradict each other, so check the current documentation from Anthropic and from the tool. The safest way to build automation is on an API key.

Is writing blog content with Claude’s help prohibited?

No. The policy does not prohibit using AI to create content. It prohibits deception about where content comes from: fake authors, media outlets and reviews, hidden sponsorship and networks of “independent” sites. The section on fraud also contains a line about plagiarising or submitting AI-assisted work “without proper permission or attribution”, which the policy does not elaborate on. The “media” category, which used to require a human and disclosure when content was published automatically, has disappeared from the list of high-risk use cases.

How do you appeal a Claude account ban?

According to Anthropic’s Help Center, log in to the banned account at claude.ai and use the appeal form, which is available only after logging in. The Safeguards team investigates the cause. Free, Pro and Max users banned for violating the Usage Policy can export their data or delete the account from that screen. Anthropic gives no response time, and the page says response times are “longer than normal”. When only an organisation is on hold, the restrictions screen has a “Request a review” option.

Does the new policy apply to the API and Claude Code?

Yes. The policy covers people who use the apps, including Claude.ai and Claude Code, developers and companies using the API, customers who use Claude through cloud providers and authorised resellers, and end users of products that have Claude inside. For the API, according to the Help Center, warnings relate to violation thresholds across the whole API account.

Is Poland on the list of supported countries?

Yes. Poland is on the list in the Supported Regions Policy effective from 12 November 2026. The policy also bans use by people who are physically located in an unsupported region, even if they work for a company from a supported country, and by entities incorporated in, headquartered in or controlled from such regions. Anthropic does not state an ownership-control threshold.

Worth remembering

  • On 12 November 2026 a new version of the Usage Policy and the Supported Regions Policy takes effect. As of 18 October, the Consumer Terms (8 October 2025) and the Commercial Terms (17 June 2025) carry the same dates as before.
  • The Usage Policy covers users of the apps (including Claude Code), the API, customers who come through cloud providers and resellers, and end users of products that run on Claude.
  • In enforcement, the threshold “if we learn” has changed to “if we suspect”, and warnings and access limits have joined the list of measures. The sentence saying that a blocked response is not evidence of a violation is new.
  • The policy itself does not describe an appeal procedure. You appeal through a form after logging in to the banned account, and Anthropic gives no response time.
  • The list of platform abuses now includes, among other things, a ban on creating multiple accounts to avoid detection, on getting around sign-up and log-in, on paying with someone else’s or fraudulent payment methods, and on reselling access through consumer subscriptions.
  • The new section on deceptive campaigns names fake reviews, hidden sponsorship, networks of “independent” sites and seeding the sources from which search engines and AI draw their answers. The line is hiding where content comes from, not using AI as such.
  • The requirements for high-risk use cases now have 11 areas and a list of exceptions. General and educational content is excluded, while personalised advice requires a human and a notice that AI was used. The “media” category is gone.
  • The ban on “sustained and needless” abuse of models is to be enforced mainly by Claude ending the conversation. The policy does not define a threshold.

Summary

On 12 November Claude’s acceptable use policy changes, not its terms of service. The changes do not overturn the rules, but they shift several things that matter: the suspicion threshold for bans, the list of platform abuses, the requirements for high-risk use cases and the new section on deceptive campaigns.

For work in SEO and marketing there are two conclusions. Accounts, automations and third-party tools are a concrete list of things to check. The harder question is whether any practice in your work relies on hiding who is behind the content. The policy will not answer that for you. It does not settle every borderline case, and the appeal procedure stays outside it. That is why it is worth knowing not only the list of prohibitions but also the route you would have to take if your account were ever banned.

Want to be sure your content and AI tools are set up honestly?

I will review how content is produced with AI in your company and show which practices are worth putting in order before they become a problem in search or with your model provider.

See the SEO audit

How this text was written

I will not pretend it was written entirely by hand, by candlelight. For the research, the proofreading and the code of this page I used a language model. It is a good tool for digging through documentation and catching the typos you no longer see on the third read of your own text. The English version was also prepared with a model, from my Polish original.

What the model did not do: it did not choose the topic, form the arguments or decide what this data means for your site. That stays with me, as does the last word on what goes into the text and what comes out of it.

Where the model helped

  • digging through sources and organising data
  • proofreading and the English version of the text
  • code, page layout and infographics

What stays with me

  • the choice of topic and what is worth saying about it
  • the arguments, the reading of the data and the conclusions
  • fact-checking and the decision on what goes into the text
Sebastian Baryś – freelance SEO specialist

Written by

Sebastian Baryś

Freelance SEO specialist · Częstochowa, Poland, and remote

I started when nobody had dreamt of Core Web Vitals and rankings were checked by hand. Today I work on making companies visible not only in Google, but wherever people ask questions. I write only about things I check myself in client projects.

SEOGoogle AdsTechnical SEOSEO in the AI era

Let’s put in order how AI works in your content

We will go through your content processes, AI tools and practices that may be risky, and agree what is worth changing before the next rule changes.

Get in touch