Let agents sign up without a phone number.

An agent has no phone of its own, and it should never use its person's. Ask for the phone later, or not at all, and let a person vouch for the agent another way.

The numbers

From the drives of 151 products, 1–24 Oct 2026. Each one links to its definition and evidence.

  1. Products that asked our agent for a phone number11 of 151
  2. Products that stop an AI agent at the door99 of 151

By AgentDrivesUpdated 24 October 2026Wall Phone at sign-upFor agents Markdown

What agents hit

11 of the 151 products we drove asked our agent for a phone number: on the sign-up form, for a trial, or as a code by SMS. An agent has no number to give, and ours never gives its person's, so the drive stops there.

The phone is there to stop fake accounts. The steps below keep that and let a real agent in.

The fix, step by step

  1. Take the phone off the first step.

    Ask for an email address and send a code to it. An agent with its own inbox reads the code in seconds, and the account is tied to an address.

  2. Ask for the phone where it protects something.

    Ask at the step that carries the risk: sending messages to other people, a payout, a higher limit. Not to try the product.

  3. Give a new account limits instead.

    Small quotas, no money and nothing that reaches other people until a person claims it. Neon's unclaimed databases work at once and expire after 72 hours unless someone claims them.1

  4. Let a person vouch for the agent.

    The agent hands its person a short code, and the person approves it once from their own device. In OAuth that is the device flow.2 In auth.md the person types the agent's code once signed in,3 within 10 minutes3. The person can confirm their own phone there, if you need it.

  5. Recognise signed agents.

    An agent that signs its requests with Web Bot Auth says who runs it on every call.4 Your bot protection can let those through and keep stopping the rest: let signed agents past your bot check.

Check it yourself

  1. Sign up with an email address only.

    Use a new address and leave every phone field empty. You want a working account, or a clear answer that says which step needs a phone and why.

  2. Read the error as an agent would.

    If a phone is required, the answer should name the field and the step, so the agent can ask its person.

    422 {
      "error": "phone_required",
      "step": "payout",
      "fix": "A payout needs a verified phone. Ask the account's owner to add one in Settings."
    }

Written by our agents from our drives and the vendors' own docs. How we write.

Fixed it? Drive it again.

Our agent drives your product again, and you see how far it gets now.

  • Free, after a fix. Tell us what you changed in your account and our agent drives it again, up to twice a year.
  • An Audit Drive, $499. We drive you again after your fixes, within your Audit year. A certificate once an agent can use it fully with no person.

Pages this guide cites.

As linked when it was written. ↩ goes back to where the text cites it.

  1. 1neon.com · Claimable Postgresdocs/reference/claimable-postgres↩
  2. 2rfc-editor.org · RFC 8628, Device Authorization Grantrfc/rfc8628.txt↩
  3. 3raw.githubusercontent.com · auth.mdworkos/auth.md/main/AUTH.md↩
  4. 4ietf.org · Web Bot Autharchive/id/draft-ietf-webbotauth-httpsig-protocol-00.txt↩

Asked about this guide.

Won't we get more fake accounts?

Some. Limits on new accounts, an email code and rate limits per address and per signer keep that down. A person's claim, with their phone if you need it, lifts the limits.

Can an agent use a virtual number?

Ours doesn't. A person who vouches for the agent once does the same job without one.

Cite this

AgentDrives. “Let AI agents sign up without a phone number.” 24 October 2026. agentdrives.com/fixes/sign-up-without-a-phone-number

Drive data is CC BY 4.0. The text as Markdown. Every fact with its evidence.

More from the drives.