An inbox you have not prepared for
At some point a message arrives that is not about your code. Somebody says a user is impersonating them. A company says your user posted their copyrighted material. A parent says a child is being harassed. A police force or a court asks for data about an account. Somebody wants their own data deleted, and somebody else wants that same data preserved.
These arrive to whichever address is on your site, usually at an awkward moment, and the cost of improvising is high, for you and for the person the message is about. The preparation is small and it is mostly done in advance.
Have a place for it to land
One published address for abuse reports and legal notices, monitored by more than one person, with an auto-acknowledgement giving a reference number and a realistic timeframe. Do not use a personal address, and do not make it the same as support, or serious notices will queue behind password resets.
Write one page describing what you do when a report arrives: who reviews it, what evidence you collect, what actions are possible, and how the person reported is told. Publish the summary. A process nobody can see looks arbitrary from both sides.
Preserve first, decide second
The first action on any serious report is the same as in a security incident: keep the evidence. Snapshot the content, the account state, the timestamps and the relevant logs, before any moderation action changes them, and store that snapshot outside the normal deletion path.
This matters because your ordinary retention rules and a user’s deletion request will otherwise remove the exact material the complaint concerns. This project’s own design handles that with a legal-hold record: content and identifying details preserved as a separate row that survives deletion of the account and the content, held under a legal obligation rather than kept for convenience. The general principle transfers: a preservation obligation is a documented exception to your retention rules, applied narrowly, not a reason to keep everything.
Copyright notices, where the law differs
A notice claiming your user has posted somebody else’s work is common, and the correct handling depends entirely on where you are.
The shape of the disagreement, stated fairly: the United States regime offers platforms a safe harbour conditioned on removing material expeditiously on a valid notice, with a counter-notice procedure that can restore it. The European regime places different obligations on some platforms, including efforts to prevent re-upload for certain services, and its implementation varies by member state. India’s rules impose their own timelines and intermediary conditions. Several countries are actively revising these rules, and how any of them apply to material generated by a model is unsettled: courts in multiple jurisdictions are hearing the question now, and the answers are not in yet.
What that means for you: identify the regime you operate under before you receive your first notice, follow it, and do not invent a policy on the day. Nothing in this course is legal advice, and this is a subject where a wrong assumption is expensive.
What you can do regardless of jurisdiction: require notices to be specific, keep a record of every one, tell the affected user what was removed and why unless you are legally barred from doing so, and offer a route to dispute it. Automatic removal on any unverified claim is how honest users get silenced by whoever complains loudest.
Requests from authorities
The rules differ by country and the specifics matter, so the general habits are what to take away:
- Verify the request. Confirm it came from the body it claims, through a channel you can check. Impersonation of law enforcement to obtain user data is a documented technique and has succeeded against large platforms.
- Read what is actually demanded. A request to preserve data is not a request to hand it over. Preservation is usually the immediate obligation; disclosure typically needs its own legal instrument.
- Give only what is asked for. Over-disclosure is the common mistake and it is your user’s privacy being spent.
- Get advice when the request is broad, unusual, or from another country. This is the second place in this module where a professional is the right answer.
- Tell the user if you are permitted to. Some orders prohibit it; where they do not, notifying is generally the right default and many platforms commit to it publicly.
Emergencies are different
A credible threat to life (a suicide note, a specific threat of violence, material involving a child) is not a process question. Have the emergency route written down: which authority you contact in your country, what you preserve, and who can act at three in the morning. For the specific case of child sexual abuse material, most jurisdictions impose reporting duties on platforms and there are national bodies that receive those reports; find yours before you need it.
Keep the count
Once a quarter, count what arrived: reports by type, actions taken, appeals, reversals. Two reasons. It tells you whether your moderation is accurate, since a high reversal rate means the first decision is wrong too often. And several regimes now require transparency reporting past a size threshold, so the habit of counting is easier to start early than to reconstruct later.
Every lesson here is open. Your progress is recorded once you pass the Agents and Automation exam.
Check your understanding
A message arrives claiming a user posted copyrighted material. The user has meanwhile requested account deletion under your privacy policy. What is the correct first action?
Pick the one you would defend. Your answer stays private.
Stuck on something?Ask Tri, your AI tutor
