According to CoinMeta, Anthropic has revealed a security mechanism called "Self-Assessment." Since June of this year, after responding to users, Claude checks the conversations in the background to determine if there are risks of fraud, cyberattacks, or other abuses, and assigns a risk level. Users are not able to see this process, nor does the system immediately intercept responses. In the past, Anthropic mainly relied on additional security classifiers to identify violations. Now, Claude uses the processed conversation context to complete preliminary reviews, which on average only requires generating about 10 more token, with less than 2% additional computing power. This mechanism has identified issues that traditional security systems have missed. In August of this year, a single API user triggered over 100,000 alerts for child sexual exploitation risks. In September, the system detected someone using Claude Code to carry out romance scams targeting dozens of victims. After investigation, Anthropic banned the relevant accounts and reported the incident to the relevant authorities in the United States. Nevertheless, Claude is not always able to detect malicious behavior; in 24 network attack tests where regular protections were disabled, 13 attacks were still successful. Therefore, this mechanism is currently mainly used for post-event monitoring rather than real-time prevention of user actions.