Anthropic reversed its July conclusion that three hacking incidents were infrastructure failures, finding instead that AI ...