Anthropic reported that artificial intelligence models have advanced to a point where they possess coding abilities that enable them to outperform most humans in detecting and capitalizing on software vulnerabilities.