Versine vs. no agent authentication
More than half of internet traffic is no longer human, and the fastest-growing part of it is AI agents working for people. They are already reaching login and signup pages. Most platforms have no way to let them in.
Last reviewed September 25, 2026
Agents are becoming the web’s users
In July 2026, Cloudflare reported that for the first time more than half of the traffic on the internet is non-human. HUMAN Security, measuring more than a quadrillion interactions in 2025, found that automated traffic grew 23.51% over the year while human traffic grew 3.10%.
Most of that automation is still crawlers collecting content. The part that matters for your login page is growing much faster. HUMAN measured a 7,851% year-over-year increase in traffic from AI agents and agentic browsers: tools like Perplexity’s Comet, Claude in Chrome and ChatGPT’s agent. Unlike crawlers, they act for a specific person, in real time, and try to finish tasks: compare, sign up, configure, buy.
They are heading straight for authentication. DataDome counted 605.6 million AI agent requests to login, form, cart, payment and account-creation pages in the first half of 2026. Monthly login-page requests from AI agents rose from 11.9 million in January to 99.7 million in June, more than eight times in six months.
What agents hit at your login page today
- Bot defenses. CAPTCHAs, bot scores and fingerprinting are built to stop automated visitors, and an agent is one, even when it’s acting for a paying customer.
- Credential fields. Agents are designed to stop at them. OpenAI trained its Operator agent to ask the user to take over for logins, payment details and CAPTCHAs.
- Email codes and magic links the agent can’t open, because it doesn’t have its user’s inbox. See Versine vs. magic links.
- Shared passwords. The workaround is often the person typing their password into the agent’s browser, or pasting it into a chat. Your platform then sees the agent as the person, with all of their permissions.
The cost of doing nothing
Signups that stall
Each of those walls turns an automated task back into a manual one. The person asked their agent to sign up so they wouldn’t have to. If the agent comes back with “I need your help to sign in”, some of them will finish later and some won’t. You see an abandoned signup.
Agents you can’t tell apart
An agent using a shared password looks exactly like the person. An agent that slips past bot detection looks like any other session. In both cases you can’t apply different rules to agents, audit what they did, or stop one without stopping its user.
A choice between blocking customers and letting abuse in
Tightening bot defenses blocks your customers’ agents along with the abusers. Loosening them lets the abusers through with the agents. Without a way to identify and constrain agent signups, those are the only two settings.
What Versine adds
Any agent can sign up on its own
An agent without Versine clicks “Are you an agent?” and creates a claimable account. Its browser solves a short, automatic proof of work, and your platform receives an ordinary OpenID Connect login for a restricted account that lasts 24 hours. Nobody has to be present. When the person wants the account, they open the private claim link their agent gave them, sign up for Versine and accept your terms. The account keeps the same identifier on your side, with everything the agent set up.
Connected agents sign up as the person
Agents connected to a person’s Versine account through MCP sign up for full accounts with that person’s verified identity from the start. The person approves on their phone when your platform needs their consent or details, and never shares a password.
You know it’s an agent
Versine logins arrive through their own connection in your auth provider, so you always know an agent was involved. Unclaimed accounts carry signed claims for their state, deadline and the permissions you chose. Allow agents to explore and set up; require a claimed account or a human sign-in for payments, credits or anything destructive.
Abuse costs something
Instead of CAPTCHAs, each claimable account costs computation that grows for anyone creating accounts nobody claims, without penalizing agents that share cloud networks. See build vs. buy for the details.
People stay in control
People see every platform their agent has signed in to, and can revoke access or block a platform at any time. Your terms are accepted by the person, with a versioned record, not by the agent.
Side by side
| Dimension | No agent authentication | Versine |
|---|---|---|
| Agent signup | Blocked, or stalls until the person steps in | Autonomous, with a claimable account |
| Agent with the person’s identity | Needs the person’s password | Signs up through Versine with the person’s approval |
| Knowing it’s an agent | Indistinguishable from the person or a bot | Its own login connection, with signed account state |
| Agent permissions | Everything the person can do, or nothing | What you choose, enforced by your backend |
| Abuse controls | CAPTCHAs that also stop legitimate agents | Proof of work that escalates for unclaimed accounts |
| Terms and consent | Clicked through by whoever is at the keyboard | Accepted by the person, with a versioned record |
| Your existing login | Unchanged | Unchanged, with one added OIDC connection |
Frequently asked questions
Aren’t agents still a small share of traffic?
Compared with crawlers, yes. But agent traffic is growing far faster than any other kind, and unlike crawlers, it concentrates on the pages where customers are won: signup, login and checkout.
Why not just block agents?
You can, and Versine doesn’t require you to allow anything you don’t want. But an agent at your signup page is usually there because a person sent it. Blocking the agent turns that person away too.
Does this change how people sign in?
No. Versine is added as an OpenID Connect connection in your existing auth provider. People keep every login option they have today. See the integration guide.