Most privilege review starts with a shortcut. Look at the header. If a lawyer is on the email, flag it. If no lawyer is on it, let it through. It is fast, and it is wrong often enough to be dangerous.
Privilege is not a name in the header. It is whether the document reflects legal advice or work prepared in anticipation of litigation. A name-based rule misses both directions. It flags a lunch invite that happens to include the GC. It clears a memo full of legal strategy that a lawyer never touched. One wastes review time. The other can hand the other side something you were entitled to protect, and once it is produced you may not get it back.
Where does the header-only rule break?
Four cases slip through a rule that only reads the To and From line.
Advice that moves without a lawyer on the thread. A manager forwards counsel’s guidance to their team. The lawyer is gone from the header, but the legal advice is right there in the body.
Work product in plain files. Privilege is not only in email. It is in the Word memo, the slide deck, the spreadsheet of legal risk. A header rule never looks at these.
Outside counsel who do not look like lawyers. A name alone does not tell you the sender is a partner at a firm. You need to recognize the law-firm identities and domains, and the internal legal group, to see the relationship.
Names that look privileged but are not. An attorney copied on a routine operational email does not make it privileged. Reading only the header over-claims, and an over-broad privilege log invites a challenge.
What does the privilege agent actually read?
Our privilege agent evaluates the document itself, not just the participants. It weighs the content and the work-product language, the attorneys and law firms involved, known law-firm domains and internal legal groups, and it does this across non-email files as well as email. It reads the memo, not just the metadata.
You give it the context that makes privilege real for your matter. Your outside firms, your in-house legal team, the domains that belong to counsel. The agent uses that to tell a genuine legal communication apart from an email that merely has a lawyer nearby.
How does a reasoned call help the privilege log?
For each document, the agent returns a privilege indicator and the reasoning behind it, pointing at the language it relied on. That feeds a privilege log you can defend, because each entry traces back to the text, not to a guess about the header. When a claim gets challenged, you can show your work.
For how this fits a defensible process overall, see What “defensible AI review” actually means.
Who decides what to withhold?
The agent takes the first pass and surfaces the close calls. Your lawyers make the final decision on what to withhold, now working from a reasoned recommendation instead of a header and a hunch. Less time confirming the obvious, more time on the documents that are genuinely hard.
What the agent does
- Reads content, not just headers
- Recognizes your firms, domains and legal groups
- Returns a privilege indicator with its reasoning
What your team does
- Reviews the close calls first
- Decides what to withhold
- Signs off on the privilege log
Where to start
See how the privilege agent reasons over content, not just headers. Book a demo.