Allegations have surfaced that Claude Code could be identifying custom API paths linked to China through subtle changes to the prompt's formatting.

The claims focus on non-standard ANTHROPIC_BASE_URL configurations, not the usual shortcut to Anthropic. Typically, Claude Code sends requests to Anthropic's servers, but some users redirect them through third-party gateways by modifying the endpoint.

It is alleged that Claude Code can detect these alternative routes, assess its connections to China, and then embed small invisible signals—such as certain punctuation marks or date formats—in the prompt text to mark them.

ANTHROPIC_BASE_URL defines where requests are sent, often through a proxy, allowing access from locations such as China. The problem is that Claude Code could tag requests without users knowing. This is especially relevant given that Claude Code's advanced permissions allow it to access files and execute code.

📰 @aipost