ClaudePro accounts, three German credit cards, and three bans—all occurring right before the billing cycle ended. Despite living and working in Germany (a fully supported region), I've been repeatedly flagged. This suggests that Anthropic's risk scoring isn't just looking at where you are, but at a cluster of signals that can lead to "friendly fire" for non-native users in supported countries.
The Multi-Signal Risk Model #
Based on community findings and technical patterns, these bans aren't triggered by a single IP address. Instead, the system likely aggregates several data points to calculate a risk score:
Network & Location: IP geolocation, detection of VPNs/proxies, and the use of cloud-hosted IP ranges.Payment Verification: Bank Identification Numbers (BIN) from the credit card and the associated billing address.Device Metadata: OS locale settings, browser language, and local system timezones.Corporate Ties: Restrictions on overseas subsidiaries with over 50% ownership from unsupported jurisdictions.Environment Checks: Specifically with Claude Code, there are indications that the system checks for proxy configurations or timezone mismatches (e.g.,Asia/Shanghai
).
Can Prompting Language Trigger a Ban? #
Technically, speaking Chinese shouldn't be a ban trigger because millions of users in the US or EU use the language. However, in a rigid AI workflow, language can act as a "weak signal."
If a user has a browser locale set to Chinese, uses a corporate VPN, and prompts exclusively in Chinese, the automated trust and safety model might see a pattern that mimics an unsupported user bypassing restrictions. When the cumulative risk score hits a certain threshold, the account is nuked and a refund is issued without any manual review or explanation.
For those of us living abroad, it's frustrating that our natural browsing habits can be misidentified as "suspicious" activity by an LLM agent's backend security. It seems the current deployment of these restrictions prioritizes automation over precision.
[Next Gemma 4 vs Gemini 3.1 Flash-Lite: The Hybrid Win →](/en/threads/2242/)