very fine line: anthropomorphise too much, and we think this is a pure alignment problem that imminently and smoothly extends to other domains. anthropomorphise too little, and we think this is exclusively a cybersecurity problem, not an alignment problem. both would be wrong!