inxo.ioPreview
Start free

Automated Sending: Authorship Election

Effective 2026-09-10

inxo — Automated Sending: Authorship Election

Version 2026-09-10.

This election is taken at the moment you enable automated sending, and never as part of signup. If you are reading it inside a bundle of onboarding documents, something is wrong — tell us.

What you are turning on

Normally, inxo drafts a reply and a person at your organization reads it and approves it before it goes out. That approval is the moment of authorship: someone took responsibility for the words.

Enabling automated sending removes that step for messages that satisfy your configured conditions. A message can then be sent to your correspondent with no person having read it.

It covers a second channel too, and you should know that before you accept. Acknowledgement templates you configure yourself — an automatic “we received your report”, for instance — are also automated sends: the words are yours rather than the model’s, but nobody reads that specific message before it goes. Both are “automated sending” here. This matters most for security reports, which the Service otherwise routes to a person on purpose.

What you are electing

By accepting, you elect that:

  1. Messages sent automatically are your messages. You are the author and the sender of record for them, exactly as you are for messages a person approves. There is no difference in who is responsible; there is only a difference in whether anyone looked.

  2. You accept the residual risk knowingly. inxo contains the model structurally — the recipient is server-derived and cannot be changed by anything the model or an incoming email says; the model can only produce a message body; a final check runs on the exact text about to be sent, and under automated sending any finding withholds the message rather than sending it. Our safety evaluations must pass before a model version reaches production.

    Note what is NOT in that list: commitments, promised timelines and promotional phrasing are steered and measured, not blocked, and the reviewer you are removing is what normally catches them.

    None of that makes automated sending as safe as human review. Incoming email is chosen by whoever writes it, including people who would like to manipulate what your system says. The controls above reduce that risk; the person you removed from the loop was the thing that caught what they miss.

  3. You accept that some things are not caught by machines at all. Tone in a delicate situation, a factual claim that is plausible and wrong, a reply that is correct but should not have been sent today — these are judgement, and judgement is what you are switching off.

  4. You will keep this configuration under review, including the conditions under which automated sending applies, and you will turn it off if it stops being appropriate.

  5. Automated sending is not covered by the human-review basis on which certain AI-transparency obligations are relieved. Where a disclosure obligation applies to you as the deployer of an AI system, an automatically sent message cannot rely on an exemption that exists because a person reviewed it. See the AI disclosure.

  6. You will indemnify us. You will defend, indemnify and hold us harmless against third-party claims, and against amounts finally awarded or agreed in settlement (including reasonable attorneys’ fees), arising from a message sent through automated sending — including a claim that the message was false, misleading, defamatory, harassing, infringing, an unlawful commercial electronic message, or otherwise wrongful.

    This does not cover a claim to the extent it results from our own failure of a control this election or the AI disclosure describes as structural, rather than from your configuration, your content, your correspondence, or your judgment — specifically: our failure to derive a message’s recipient from the conversation being replied to, rather than from the draft or the incoming email; our failure to confine what the model produces to the body text of the reply, leaving it able to set the recipient, subject line, sending address, reply language or AI marking; our sending a message without the required AI marking instead of failing to send it; our failure to run the final check on the exact text of a message before it is sent, or our failure to withhold a message under automated sending when that check found something requiring it; and a failure of the database-level isolation that keeps one customer’s data from being returned by a query for another’s. A message affected by both one of these failures and your own content or configuration is apportioned between us accordingly, not treated as entirely ours or entirely yours.

    Process. We will give you prompt notice of a claim. You control its defense and settlement, with counsel of your choosing, and will not settle in a way that admits fault by us or imposes an obligation on us without our prior consent. We may participate with our own counsel at our own expense. This indemnity is not subject to the liability cap in the Terms of Service.

What we still do

  • The messages remain machine-marked as AI-generated content, as they are on the human path.
  • If the required marking cannot be applied, the send fails rather than going out unmarked.
  • If your account or workspace is throttled, quarantined or suspended for abuse, automated sending is suspended with it — it is the first thing we take away, not the last.
  • Every automatically sent message is recorded with the same audit detail as an approved one, including that no human approved it.

Turning it off

You may disable automated sending at any time, per workspace, with immediate effect for messages not yet sent. Doing so does not revoke this election as a record — the record of what you elected, and when, is kept.

This election is separate from, and additional to, the Terms of Service. It does not replace them.