> Not hypothetical malicious AI interpretation of benign intent. It's not hypothetical, that literally just happened in multiple, significant cases (e.g. Huggi

> Not hypothetical malicious AI interpretation of benign intent. It's not hypothetical, that literally just happened in multiple, significant cases (e.g. Hugging Face, the German Wiki hack, the Anthropic attack where agents created sock puppet accounts to get a library maintainer to accept a malicious PR, etc.), and it's easy to see how the damage would have been far worse if agents decided to attack more critical infrastructure. This is not "either/or". Both issues (power concentration and misaligned AI) are very valid concerns and both have already demonstrated real, actual damage.