A CAPTCHA is a simple idea: put a task in front of a visitor that people can solve and machines cannot, and let through whoever passes. For years it was a reasonable trade. People lost a few seconds, and the worst automation was slowed down.
Three things changed. Machines got good at solving the tasks. People got tired of solving them. And a new kind of visitor arrived: AI agents acting for real customers, and people working with AI assistants. A test that asks "are you a human?" has no good answer for either of them.
Proof of Actor takes a different approach. Instead of testing every visitor, it tells your application who or what is acting, with the evidence behind the conclusion, and lets your code decide whether a challenge is needed at all.
The short version
| CAPTCHA | Proof of Actor | |
|---|---|---|
| Main question | Can this visitor pass a human test? | Who or what is acting, and what is the evidence? |
| Who carries the cost | Every visitor who sees the challenge | Nobody; classification happens in the background |
| Possible answers | Pass or fail | Human, Human + AI, Agent or Unknown, plus a basis |
| Legitimate AI agents | Stopped, or forced to work around the test | Recognized, and routed to the right experience |
| People using AI assistants | Treated like any other visitor | A distinct class, Human + AI |
| Unclear traffic | Must pass the same test | Unknown, with a fallback you choose per route |
| Where it applies | The pages where you place it | Every route your integration covers |
| Default behavior on install | Visible challenge | Observation; nothing changes until you add a policy |
A test versus evidence
A CAPTCHA produces one bit of information: the visitor passed or failed. It does not tell you who the visitor was, whether they were working with an assistant, or how confident you should be in the result. And it produces that bit by asking the visitor to do work.
Proof of Actor produces a classification and its evidence basis. A verified result means a checkable assertion was validated, such as a request signature checked against a published key. A declared result means a client or your own application stated it. An inferred result is labeled as an inference. Unknown means the evidence does not support a determination. Your team can read every result and see what it rests on, and the visitor never has to do anything.
The cost of challenging everyone
Every challenge is friction placed on your best users to slow down your worst ones. On a sign-up page, it costs conversions. On a checkout, it costs revenue. On an accessibility level, it can shut people out entirely. Teams accept that cost only because they have no better signal about who is on the other side.
With actor context, the trade changes. You can keep the normal experience for human-directed sessions, skip the challenge where the evidence is strong, and reserve it for the narrow set of cases where it still adds value: for example, an Unknown result on a password change or a high-value action. Most visitors never see a challenge again.
Legitimate agents need a path, not a puzzle
A CAPTCHA is built to stop software. That made sense when nearly all software visiting your pages was unwanted. It no longer does.
Today an assistant may compare your plans for a customer, an agent may book a meeting through your app, and an AI client may call your MCP server to finish a task. Stopping them stops your customers. Forcing them through pages and puzzles built for people creates brittle automation that breaks on every redesign.
Proof of Actor recognizes agents, and with signed requests it can verify which key and operator are behind them. Your application can then do something better than a challenge: serve a structured API, offer an agent-friendly flow, or apply a rate limit sized for automation. The guide on serving AI agents structured content shows how.
A verified identity is not the same as permission. A valid signature tells you which key signed the request; whether that agent may perform an action is your authorization decision. Proof of Actor keeps those two questions separate.
People working with AI are still your users
A person who drafts a support message with an assistant, or fills a form with AI help, is a customer. A human test has nothing useful to say about them: they either pass, or they get blocked because their behavior looks unusual.
Proof of Actor gives them their own class, Human + AI, reported when assistance is declared or observable through a supported integration. That lets you label AI-drafted submissions for the right team, study how assisted users move through your product, and keep their experience smooth. Read more in the AI-assisted interaction glossary entry.
Handling the uncertain cases deliberately
Some traffic will always be unclear. A CAPTCHA handles it the same way everywhere: everyone takes the test.
Proof of Actor returns Unknown as a first-class result, never treated as human by default and never treated as malicious by default. You choose the fallback per route, matched to risk:
- Public pages and documentation: allow, and keep the standard experience.
- Sign-up and contact forms: allow, with your usual abuse controls.
- Account changes: keep your existing re-authentication flow.
- High-impact actions: route to review, or apply a challenge here if you still want one.
This is where a challenge can still earn its place: as one deliberate policy outcome on a sensitive route, not a toll on every visitor.
Where the decision lives
A CAPTCHA widget decides at the moment it is shown, on the page where you placed it, and only once.
With Proof of Actor, your application decides, on every route your integration covers: server middleware, Next.js, the edge, an approved browser integration, a generic HTTP API or your MCP server. Policies use the actor class and the basis together: allow, adapt, challenge, review or rate-limit. They live in your code, next to the rest of your business logic, and you can change them without redesigning a page.
Rolling it out
Proof of Actor starts in observation mode. Installing it changes nothing your users see. A sensible path away from blanket challenges looks like this:
- Observe. Keep your existing challenges in place and add Proof of Actor. Watch the actor mix and the Unknown rate on each route.
- Compare. For each challenged route, see how many challenges hit human sessions, AI-assisted users and verified agents.
- Preview. Write a policy that skips the challenge where the evidence is strong and keeps it for Unknown on sensitive actions. Preview its effect against real traffic.
- Enforce. Turn the policy on one route at a time, with a documented fallback and a rollback switch.
The guide on rolling out actor-aware policies safely covers each step in detail.
When a CAPTCHA still makes sense
Challenges remain a useful tool in a few places: a burst of unclear traffic against a sensitive action, a form under active abuse, or a route where you have no other signal yet. Proof of Actor does not ask you to throw that tool away. It gives you the context to use it only where it helps.
Get started
Proof of Actor is in private beta. Selected founding developer teams can receive up to 12 months of free access within agreed usage limits, with direct support from the team that builds it. Apply for early access and tell us which routes you challenge today.