As you read your email, our phishing filter is scanning it in real-time for just two things: known malicious links and language patterns that indicate a high likelihood of malicious intent. Once the scan has finished we feed the scan results back into our machine learning training model to tune our detection paradigm. There is no way to reconstruct an email from a scan result and the email is not retained in any of our systems.
The only time we save an email is when you press the button to report phishing or report an email as safe or unsafe. In that case, we hold onto the email to analyze the malicious actor's attack pattern and build defenses around their behavior. If we decide the email is not malicious, we delete it from our systems.