AI Moderation

1. Why we moderate

Email is the main route for fraud in India. Fake bank alerts, KYC and OTP requests, parcel and tax notices, fake job offers and lookalike websites reach people every day. Indo Mail checks incoming mail to protect you from these attacks and to keep our service trusted by other mail providers.

This page explains what the automated checks do, what they cannot do, and how you can challenge a result. It covers mail delivery, the scam warning banner, sending limits and reports made by users. It does not describe how we read your mail for any other purpose, because we do not.

2. What is checked, and what is not

Checks run on the message as it arrives, on our servers in Mumbai. Only the message itself is examined. No message text, attachment or address of yours is sent to an outside service. To check reputation, the server looks up only the sending server's IP address, the sender's domain and the domains of links in public blocklists (Spamhaus, SpamCop, PSBL, Mailspike and SURBL) through the internet's DNS system, and the sender domain's registration date in the public RDAP registry.

We do not read your mail to build a profile, to sell data or to target ads. We do not train models on your mail.

• Checked: sender identity, sending server, headers, links and the words they show, attachment names and file types, the words of the message (scored by the content model and a fixed list of risky phrases, then discarded).
• Not used: your contacts, your reading habits, your location, or the content of other people's mailboxes. One signal can only lower a score: if you have written to the sender before, the message is less likely to be treated as spam.
• Mail you send is checked for abuse patterns such as high volume to many recipients, as explained in our Acceptable Use rules.

3. Automated detection live today

Detection uses the Indo Shield filter (release 1.6): fixed rules, standard internet authentication, public reputation lists and one content model. These are written and tested by our engineers. Each layer adds or removes points; the total gives a score from 0 to 100. At or above your threshold (80 by default, adjustable in Settings → Spam protection) the message goes to Spam. Every warning and every message in Spam shows its reasons in plain words, and "See all 7 checks" shows each layer.

Warnings come at two levels: caution (one strong sign) and high (two or more signs, a risky attachment, or a lookalike or imitation domain). A warning never deletes your mail. You can open the message, mark it as not spam, or press "Report scam".

4. What the checks cannot do

No filter is perfect. Rules can miss a new kind of scam, and they can warn about a harmless message that happens to look suspicious, for example a real bank alert from a partner domain. We say so in the banner. If you are unsure, do not click links or share codes; contact the organisation through its official website or phone number.

Rules work best on English and on the most common Indian-language scam phrases. Coverage in other languages is limited today and is a priority for the next stage.

5. Planned: machine learning

Release 1.6 added the first learned model, the content model in Layer 5, trained only on example messages written by Indo Mail. We plan to add more learned models beside the rules. The plan is described here so that you know what to expect.

• A classifier trained on public and licensed scam datasets, and on messages that users choose to report. Reported messages are used only after the reporter agrees and after personal details are removed.
• Language models for Hindi, Bengali, Tamil, Telugu, Marathi, Gujarati, Kannada, Malayalam, Punjabi and Urdu scam wording, including Roman-script writing such as Hinglish.
• Image checks for text hidden inside pictures, a common way to avoid text filters.
• Reputation signals from the wider internet, such as known bad domains (live since release 1.6 through public blocklists).
• Models will run in India and will see only what the rules see today. They will not train on private mail.

A new model goes live only after it passes the accuracy tests in section 8 and a staged release to a small share of traffic.

6. Human review

People stay in charge of decisions that matter.

• Automated checks may place a message in Spam or show a banner. They do not close accounts.
• Closing, suspending or limiting an account needs review by a trained member of the abuse team, who looks at the evidence and the account's history.
• Reports made with "Report scam" or sent to abuse@indoemail.in are read by people. The reviewer sees the reported message only, not the rest of the mailbox.
• Content involving a child, or non-consensual intimate imagery, is escalated at once and acted on within 24 hours, with reports made to the authorities as the law requires.
• Reviewers are trained, bound by confidentiality, and checked on a sample basis by a second reviewer.

7. Appeals and corrections

If we limit your account or block your sending, we tell you why in plain words, unless the law or an investigation stops us.

• Appeal by replying to our notice or writing to support@indoemail.in. A different reviewer from the first looks at it.
• We answer within 3 working days for sending limits and within 7 days for other cases.
• If you are still unhappy, write to grievance@indoemail.in. The Grievance Officer acknowledges within 24 hours and resolves within 7 days, as the Information Technology (Intermediary Guidelines and Digital Media Ethics Code) Rules, 2021 (as amended in February 2026) require.
• If a message was wrongly marked as spam, use "Not spam". We use these signals to fix our rules. Not spam and Report spam record only the sender, the sender's domain, the sending server and link domains, never the words of the message.
• If you run a domain and your mail is wrongly blocked, write to abuse@indoemail.in with the sending domain and examples.

8. Accuracy targets

We set targets and measure them every month on a test set of real-world scam and normal messages. Targets for the learned models, which are planned, are below. Rules are measured the same way today.

We publish these figures in our transparency report once we have enough volume to make them meaningful. We do not claim results we have not measured.

9. Your responsibility

• Never share an OTP, PIN or password with anyone, even someone who says they are from Indo Mail.
• Report suspicious mail with "Report scam".
• Turn on 2-Step Verification, so that a stolen password is not enough to enter your account.
• Check the sign-in history in Settings after any warning.

10. Contact

Report abuse: abuse@indoemail.in. Appeals and help: support@indoemail.in. Privacy questions: privacy@indoemail.in. Complaints: grievance@indoemail.in.

This document is governed by the laws of India.