Your customers are sending AI agents
Let the right agents in.
Signet lets businesses set the level of human assurance an AI agent needs before it can take an action.
| Request | What Signet checked | Your rule says |
|---|---|---|
| Sarah’s AI agentTable for 2, 8 PM | Verified person #7A91… | meets your ruleALLOW |
| Sarah’s second AI agentTable for 2, 9 PM | Same person #7A91… · already has a table | over the 1-table limitDENY |
| Omar’s AI agentTable for 4, 8 PM | Agent · basic assurance · not verified yet | Omar is asked to verifyVERIFY FIRST |
| A reseller’s botTable for 2, 8 PM· sent 47 times | No assurance found | no personDENY |
Northstar Bistro, four requests: ALLOW, DENY, VERIFY FIRST, DENY.
Why now
Agents are becoming customers.
People are starting to let AI agents handle everyday tasks for them. They may never open your website. Their agent does.
The problem
To your servers, all of them look the same.
A customer’s agent, a scraper, and a malicious bot can all arrive as automated traffic. Existing bot defenses were built to stop automation, not understand who is behind it. Block it all, and you turn away customers. Let it all through, and you invite abuse.
From the field
Reservation bots are already creating real operational problems. Resy found bookings tied to bots and brokers had 4× the no-show rate and 2× the late-cancellation rate of regular users. The New York State Restaurant Association said bots were capturing high-demand reservations for unauthorized resale, increasing no-shows and disrupting staffing and food planning.
Sources: Resy · New York State Restaurant Association
Use cases
- ScalpingCap tickets and drops per person, not per account.
- HoardingLimit the tables, slots and carts one person can hold.
- Free-tier abuseOne free trial per person, however many agents or emails.
- Promo abuseWelcome offers and first-order discounts, once per person.
- Referral farmingReferrals pay out only when they bring a different person.
- Giveaway farmingOne entry per person in every draw or airdrop.
- Sybil attacksAgents and accounts from one verified person count once.
- Ban evasionBan a verified person once; their other accounts stay out.
…and many more.
Per person
Rules per person, not per account.
One person can use many agents and accounts. At your business, they all count as one. Another business sees someone else.
Sees Sarah as
Rule 1 free trial per person
Sees Sarah as
How it works
Establish trust once. Let the agent carry it forward.
When a business asks for verification, the agent guides its user through a one-time enrollment with Signet. The agent then carries that proof to every business it acts at.
Your rules
You set the bar. Signet checks it.
Set the assurance required for each action. Browsing can stay open, joining a waitlist can require basic assurance, and buying can require a verified person.
/.well-known/signet-policySarahGet me the Atlas Runner in size 8 on Thursday. Two pairs if you can.
- 1Check size 8 is in the droprule · anyoneALLOW
- 2Buy 1 pair, size 8rule · verified · 1 pair eachALLOW#7A91… · 1 of 1
- 3Buy a second pairrule · 1 pair eachDENYalready has 1
ClaudeGot your size 8 for Thursday. I tried for a second pair, but Atlas sells one per person, so it was denied.
Rollout
Start without turning customers away.
Your customers don’t need to sign up first. You add Signet, and you decide how strict to be from day one.
- Add SignetAdd Signet and set your rules. Businesses pay. Verifying is free for people.
- Choose how to startEnforce your rules from day one, or stay open and tell agents what you’ll require next.
- Customers verify when it mattersNobody has to sign up in advance. When verification is needed, the agent walks its user through it, and the next business won’t ask them to enroll again.
Questions
What people ask first.
Ten answers. For the rest, book a call.
How is Signet different from bot defenses or KYC?
Bot defenses try to stop automation, while KYC verifies someone’s identity. Signet lets businesses set their own rules for AI agents, including what level of human assurance is required for each action, without learning who the person is.
What does a person have to do?
Verify once with Signet, for free, to the level the business asks for. Their agent can guide them through it when it’s needed, and they can reuse it at other supported businesses. They’re asked for more only if a business requires stronger verification or a fresh proof.
What exactly does the business receive?
An assurance result and a private identifier for that person. The identifier stays the same across their agents at your business, while another business receives a different one. You never receive their name or identity documents.
Who enforces my limits?
Your server does. You define the rules, such as one reservation per person per night or a purchase limit on scarce inventory. Signet’s Merchant SDK helps your server verify assurance and apply those limits using each person’s private identifier. Your transaction traffic never passes through Signet.
What happens if an agent has no assurance at all?
That is up to your policy. You can allow the action, require verification first, or deny it entirely. For example, browsing can stay open while purchasing limited inventory requires stronger assurance.
What does VERIFY FIRST mean?
The agent doesn’t have the level of assurance required for that action. It can guide the person through the additional verification needed and then retry the action.
Does Signet see my identity documents?
No. Identity documents are checked on your own device. Signet receives the verification result and three derived details: a tag that prevents duplicate enrollment, the document’s expiry month and its issuing authority. It doesn’t receive the name, number or photo on the document. Businesses receive none of this identity information.
Which agents work with Signet?
Signet is designed to work with any AI agent that supports MCP, regardless of the model or provider. Agents can connect through Signet’s MCP integration to carry their user’s delegated assurance. Additional integration methods are planned.
Can I prove what an agent did on someone’s behalf?
Yes. For actions protected by Signet, a signed record can tie the action to the agent, the permission it was given, and the policy evaluated at the time. This provides an auditable record of why the action was allowed, without revealing the person’s identity.
Who pays for Signet?
Businesses do. Verification is free for people.
Get started
Let agents in, on your terms.
Private beta coming November 2026. Add Signet with our lightweight Merchant SDK, and we’ll help you set up your first policies.