Anthropic has begun limiting access to its latest AI system, Claude Mythos Preview, after early testing showed it could uncover thousands of critical vulnerabilities across widely used software environments. The model, described as general purpose, identified critical vulnerabilities spanning major…