Claude has recently triggered a shocking wave of account suspensions. Throughout the holiday period, discussions about its account bans even broke into the millions of mentions, causing a stir among AI developers in the community.

This fire was reignited by an old report. Anthropic released a report in July, which was recently unearthed and widely circulated online, revealing astonishing figures: In the first half of 2026, within just six months, Claude suspended a total of 11.4 million accounts, while only 42,000 out of 398,000 appeals were overturned.

Breaking down this number, it means nearly 190,000 accounts were deleted each month, and over 60,000 accounts were physically erased every day in the cyber world. In terms of success rate, about nine out of ten people who were banned faced a cold execution—appeals seemed almost meaningless. Banning an account is just a momentary decision by the platform, but users face a long and tortuous appeal battle to get their accounts back.

Anthropic did not release the full distribution of suspension reasons to the public, but clearly listed several scenarios that would trigger a ban. A significant portion of the report was devoted to the review of extreme violations: In just the first half of the year, it submitted over 15,000 reports of violations to international organizations such as NCMEC, using not only traditional hash fingerprint matching methods, but also fully launching its self-developed new machine learning classifier. Any instructions involving illegal abuse, privacy invasion, malicious guidance, or extreme self-harm were immediately terminated by the model's built-in review system—once a prompt touched a sensitive boundary, the system gave no second warning and directly imposed a permanent ban.

The other major area of concern was the indiscriminate scanning of large numbers of black market accounts, compounded with regional restrictions, leading many to fall at this checkpoint.