Anthropic Bans 11.4 Million Accounts in First Half of Year, Only 42,000 Appeals Overturned
Anthropic disclosed in its transparency report that it banned 11.4 million accounts from January to June 2026, during which it received 398,000 appeals, of which only 42,000 were overturned.
The reasons for bans include violations of usage policies, terms of service, or regional support policies; the Safeguards team implements warnings, suspensions, or terminations through detection and monitoring, and banned users can submit appeals for review.
During the same period, in terms of child safety, 15,079 reports were submitted to the National Center for Missing and Exploited Children (NCMEC), including 12,614 known materials identified through hash matching, 2,018 previously unconfirmed or unhashed materials identified through machine learning classifiers, and 193 other reports of child sexual abuse and exploitation risks.
The report also covers government data requests, responses to user self-harm risks, and monitoring of malicious use; the company states that the relevant data aims to support public discussions on AI platform safety.
This data juxtaposes large-scale account terminations with limited appeal successes, indicating a significant expansion in enforcement scale; capital and attention are directed towards platform safety and compliance investments, benefiting providers that can maintain policy enforcement capabilities while putting pressure on heavy users who rely on single accounts and have low appeal success rates.
Recently, Anthropic also updated its usage policies, adding restrictions on unnecessary abuse or cruel behavior by models and integrating rules related to deceptive activities.
Source: Public Information
ABAB AI Insight
Anthropic continues to publicly disclose transparency data, using account bans, appeal outcomes, and child safety reports as quantitative indicators of platform safety enforcement; the scale of bans in the first half of the year has significantly increased from previous levels while maintaining systematic reporting to NCMEC.
In terms of capital pathways, resources are invested in detection, monitoring, and classifiers to enforce usage policies and regional restrictions, while the appeal mechanism addresses some misjudgments; the motivation is to meet regulatory transparency requirements and reduce malicious use and child-related risks, while adjusting policy boundaries after improving model capabilities.
Similar cases include other large platforms responding to abuse through large-scale account terminations and hash/classifier detection; the current phase is transitioning from growth prioritization in AI services to stricter policy enforcement and transparency disclosure.
Essentially, this represents a strengthening of regulation and platform control: access rights are re-priced from the user side to permissions that can be terminated by the platform at any time, shifting value from mere availability to compliance enforcement and risk control capabilities; the limited appeal success rate makes enforcement more reliant on front-end detection rather than post-correction.
ABAB News · Cognitive Law
Banning 11.4 million accounts in six months, with only 42,000 successful appeals, shows that the scale of enforcement far exceeds the scale of corrections.
Hash matching and machine learning classifiers together send both known and unknown risks into the reporting system.
After the public disclosure of platform safety data, access rights themselves become a resource that can be terminated on a large scale.