They found a way to inject text into a google email notification (by setting the name of their google workspace account to the phishing message), and then set up a mail forwarding service to redirect the notification to the victim accounts. That way the victims receive a legit email from google but the text of the email is attacker-controlled and can point the victim to their phishing site.
It's not really a vulnerability in DKIM. The bug is in google's use of attacker-controlled text fields in their notification emails.