Insight
Is AI Email Automation Safe? Questions to Ask About Security and Privacy

Margaret Jhoy

Email holds a great deal of what an organization would least want to lose: client advice, personal stories, contracts, medical or financial details, and the ordinary back-and-forth that adds up to a confidential record. Any tool that works inside that inbox has to be held to a high standard.
The question is not whether AI can be used with email. It is whether a particular tool handles that responsibility with care.
Law firms feel this pressure acutely, though the same concerns matter for legal aid groups and for support teams that handle sensitive customer information. In a May 2026 Bloomberg Law article, Law Firms Must Embed Cybersecurity in Governance to Protect Data, attorneys Sid Mody and Randall Edwards noted that law firms are attractive targets for breaches because of the volume and sensitivity of their data. They wrote that lawyers may have ethical and legal obligations to safeguard client information, and that firms should treat vendor diligence and a security-conscious culture as part of client service, including clear policies for AI use.
A more recent commentary in Bloomberg Law, Before Approving AI, General Counsel Must Know Where Data Lives, published in November 2026, put the threshold question plainly. Before approving a tool, general counsel should know where the data is, who can reach it, and what actually stops an AI system from going somewhere it should not. A vague assurance is not an answer.
Questions worth asking a vendor
Where does our email data live, and who can access it? You should get a specific answer, not a general promise. If the vendor cannot explain this clearly, that is a reason to pause.
Are our files or messages used to train models? For confidential work, the answer should be no, and it should be in writing. A tool can be useful without learning from your clients’ or customers’ information.
What independent security standards does the vendor meet? Ask about SOC 2 Type I at a minimum, and about HIPAA alignment if your work involves health information. Certificates are not the whole story, but a vendor that cannot discuss them is not ready for this kind of work.
What does a person still control? In particular, can a message be sent without a human being approving it? For sensitive email, the send button should stay with a person.
How are incidents handled, and how are subcontractors reviewed? As Mody and Edwards argue, diligence should not stop at your own walls. It should include the vendors who touch your data.
Support teams and nonprofits that handle personal information should ask the same questions, even if they are not law firms. Confidentiality is not only a professional rule. It is also a promise to the people who wrote in.
What a careful setup looks like
A careful assistant lives in your email, works around the clock, and still leaves every outgoing message subject to review. Nothing goes out on its own.
Bruce is built around that standard. It is SOC 2 Type I compliant and HIPAA aligned, and files are never used to train models. Nothing goes out without a person’s approval.
Safety in this area is less about a single feature than about a series of clear answers. If a vendor can tell you where the data lives, what it is not used for, what standards have been met, and how a person stays in control, you can have a serious conversation. If those answers are fuzzy, it is reasonable to wait.
If you would like to talk through these questions in the context of your team, you can book a demo.
