Abstract:
Recently, Claude has been crazy and banned a large number of accounts. Developers in the AI circle directly broke the defense collectively and shouted, "The sky is falling." During the entire holiday period, discussions related to "Claude's account ban" even exceeded one million.
At this moment, a report published by Anthropic in July was uncovered and blocked on the entire network.
The numbers inside are shocking.
In the first half of 2026, in just six months, Claude banned a total of 11.4 million accounts.
Of the 398,000 appeals submitted, only 42,000 were ultimately successful.
Calculated, out of every 10 people, 9 people were "coldly executed". It can be said that the appeal is almost a decoration...
What’s more, Anthropic has just updated its Claude usage policy, which explicitly prohibits insulting and cruel treatment of Claude.
11.4 million accounts blocked in half a year
Looking back, in the past, in order to grab users and encircle territory, the major AI giants basically turned a blind eye to risk control.
But the mid-year report just released by Anthropic has directly lifted the fig leaf.
On July 23, Anthropic updated a security and trust report in its Transparency Center.
There is a special column - "Banned Accounts", which clearly records how many accounts they have banned and how many people are complaining.
Let’s break down this set of data to see how exaggerated it is.
From January to June 2026, 11.4 million accounts worldwide evaporated in half a year.
This is equivalent to an average of nearly 1.9 million accounts being wiped out every month.
More than 60,000 accounts are physically wiped out in the cyber world every day.
In the face of ban decisions, 398,000 appeals were submitted.
The vast majority of nearly 400,000 people who are willing to spend time writing complaint letters, finding customer service, and even attaching proof of employment are high-frequency heavy users and developers who truly use Claude as a productivity tool.
However, only 42,000 survived, with a success rate of 10.5%.
In other words, the vast majority of appeals are "ineffective struggles." This also shows that Anthropic’s risk control judgment mechanism is almost a “one-vote veto system” at the execution level.
I would rather kill ten thousand by mistake than let one person go.
|
|
Such a disparity in data directly pushed Claude’s account ban logic to the forefront.
After all, account bans only take a moment, and the platform can just say a word; but if users want to make a comeback, they face a long and torturous tug-of-war of appeals.
Touch the red line and die suddenly
11.4 million accounts, what mistakes did they make?
Anthropic did not disclose the complete distribution of reasons for bans, but it clearly listed several types of situations that may trigger account bans.
In the details of the report, a whole core space is reserved for the review of extreme violations of content.
The report shows that in the first half of the year alone, it submitted more than 15,000 violation reports to relevant international security organizations (such as NCMEC). It not only used traditional hash fingerprint matching, but also comprehensively launched a self-developed "new machine learning classifier".
For instructions involving illegal abuse, invasion of privacy, malicious induction, extreme self-harm, etc., Anthropic's model's built-in review is almost "death upon triggering".
As long as the Prompt touches the sensitive boundary, the system will not even give a second warning and will be permanently banned.
In addition, there is also a large number of illegal accounts that shoot indiscriminately. Coupled with the "supported area" review, many people tend to hang up here in large numbers.
No way to appeal
AI workflow returns to zero overnight
If you had blocked an account two years ago, the most everyone would have said was, "No, no, let's chat in another company."
But now, everything has changed.
Especially this year, Agent workflow has become popular and coding capabilities have exploded. Countless creators, independent developers, and scientific researchers have staked their fortunes and lives on it.
For full-stack programmers, Claude is the ace architect who can chew tens of thousands of lines of shit all night long, make up single tests, and fix emergency bugs. Scientific researchers and analysts read hundreds of documents, derive mathematical logic, and clean complex data. It is also carrying the banner.
This kind of productivity cliff that instantly returns to zero is devastating.
The funniest thing is that the path to appeal is basically dead.
Many people complained,
The appeal process is just a Google form, and after filling it out, it seems like nothing has happened.
Some people appealed three times, but were rejected all times without even a specific reason.
A full-stack developer wrote in the community: "The entire team used Claude to refactor code, built more than a dozen special projects, and accumulated tens of thousands of lines of system architecture and iteration records that have undergone multiple rounds of fine-tuning.
As soon as it went online on Monday morning, the interface directly showed that it was unavailable. I submitted my business qualification, development certificate and appeal letter according to the procedures, but what I got was a template rejection email with no logical explanation.
Two weeks of hard work came to zero in an instant.
A few days ago, a developer recorded his complaint experience in detail on GitHub:
The paid Max package was disabled without warning on October 3, and the appeal was rejected. The reason was only a general "violation of usage policy" without citing any specific terms.
After additional appeal, the reply received was, "This is our final decision."
What is even more absurd is that once an account is flagged, the appeal page will be permanently stuck on the static prompt "Your account will remain closed" without any form input box or appeal entry.
All appeal tickets sent to the official support email were automatically closed by automated robots.
What’s even more outrageous is that someone was blocked just 15 minutes after registering. The account was just created, not a single message was sent, not an API was adjusted, and it just disappeared.
What is even more disturbing is that this purge may not be over yet.
Anthropic announced in its latest policy update that new restrictions will take effect on November 12, 2026, which will clearly include a variety of situations within the scope of prohibited use.
The policy has not yet taken effect, and some people in the community have already predicted that there will be another round of battle royale.
Never bet on a Claude
Some netizens summed up a lesson learned through blood and tears, "Never bet heavily on an AI."
Someone on GitHub has built a special warehouse called "Claude Account Banning Self-Rescue Guide" to continuously track each round of account bans.
In the warehouse, a core anti-blocking principle was compiled. But at the end of the day, these are just "getting by" tips.
Faced with an automated risk control system that may swing a knife at any time, no account is absolutely safe.
In the end, this is the most ironic thing about this matter.
Every day we marvel at how much artificial intelligence has replaced humans and how strong logic it has evolved, but we often forget the most primitive thing——
In this ecosystem, we have no right to say anything about our "productivity assets."
In the eyes of the other party, the Agent that I stayed up late to feed and the dozens of workflows that I tuned are just a series of Boolean values that can be set to false at any time.
Really, don’t bet your life and fortune on one Claude.
Comments