[ihsw]

As contradictory as it sounds, they (Anthropic) are probably trying to dance the fine line where its public models can write secure code but cannot exploit insecure code.