Mindgard finds the trouble your AI talks itself into

An AI assistant can refuse a dangerous request and still leak a secret through its tools. Mindgard turns that awkward distinction into a business: finding the paths from helpful behavior to harmful action.