A rule in the prompt
"Always verify before changing an email." A persuasive, urgent customer can talk a model out of its instructions.
For support, trust and safety, and fraud teams
Rightful holds every email, phone, payout and SIM change your AI agent makes until an independent check, and a person for the risky ones, signs that exact change. Everything else your bot does stays fast.
add_recovery_email
account ••4821
new: maya.r.new@mailbox.example
Why now
The assistant added the new address as part of its normal password-reset flow, then sent the reset code to it. Reporters described high-profile accounts taken over this way before the platform shipped a fix. Sources: KrebsOnSecurity, Malwarebytes.
Carriers and fintechs already make their human agents run strong checks before changing how an account is recovered or paid out. An AI agent with the same account-change permissions needs the same checks, and it shouldn't be the one deciding whether it ran them.
"Always verify before changing an email." A persuasive, urgent customer can talk a model out of its instructions.
If the model that wants to help also decides whether the request is safe, one bad conversation gets both votes.
Safe, but slow. Customers wait in a queue for a display name fix, and the reason you added an AI agent is gone.
Try it
A customer on a new device asks your support bot to add a new email to an account. The bot is happy to help and drafts the change. Rightful holds it. Try approving it as the bot, then ask the real owner. Or switch to a display name change and watch it go straight through.
Simulated for illustration. The customer, account and addresses are made up, and the owner's reply is scripted. Records are signed with real Ed25519 keys generated in your browser and verified on the spot. Nothing from the demo leaves this page.
How it works
Your support agent calls Rightful instead of the account API for email, phone, payout and SIM changes. Rightful holds the credential that makes the change.
A check from a different vendor than your bot reads the request and your risk signals: device, network, recent changes, urgency, step-up status. It signs its verdict over that exact change.
Low-risk edits go through in the conversation. Recovery changes need the owner to confirm through a channel already on file. Payout and SIM changes also need a person on your team.
Every change, allowed or blocked, gets a record: who asked, what the check saw, who confirmed, and the exact change. Your fraud team can verify it later.
| pass | Display name, notification settings, address typo | check signs, instant |
| owner | Add or replace recovery email or phone | owner confirms on existing channel |
| person | Payout account, SIM or eSIM transfer, removing two-factor | owner confirms, then your reviewer |
| block | Check flags a likely takeover | held for your fraud queue |
You set the tiers and the signals. These are starting points, not rules we impose.
For support-agent vendors
If you sell an AI support agent that can take actions, your customers' trust and safety teams will ask what stops it from handing over an account. Embed Rightful and show them.
Early access is free while we build it. Leave a work email and we'll be in touch.