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 inse

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.