Report an issue

Practice quizzes for Australian citizenship, driving, and immigration topics

Quizzes

It could expose false or harmful tool results, attempt to manipulate the model's behaviour via crafted responses, or exfiltrate data passed to it

It could directly modify the weights of the AI model to change its future behaviour permanently

It could block other MCP servers from connecting by consuming all available network ports

It could charge the user's Anthropic account for extra API credits without authorisation