Access policy

A site owner's access policy, drawn as a flow, that rewards visitors for declaring whether they are a person or a bot and rate-limits or blocks those that stay silent.

The problem

Today a site can mostly allow traffic, block it, or challenge it. Increasingly, bots and automation are being treated as hostile to the web and websites. An agent that doesn't say what it is looks the same as one that does, and so bot-ness is inferred from behavior rather than an explicit declaration. In cases where bots or automation are blocked, declaration is disincentivized. For site-holders, the drawbacks (e.g. declincing real user visits) have begun to outweigh the benefits (e.g. discovery by services like Google or ChatGPT) of allowing any bots, crawlers or other types.

The idea

Incentivized declaration. A visitor that declares what it is, a person or a bot, gets more than one that stays silent. An undisclosed visitor is heavily rate-limited or blocked, which is cheap for the site to do. A disclosed one gets access shaped to what it is.

The owner sets this up as a simple flow, like a workflow builder, and the rules are enforced at the edge. The blanket rate limiting aims to incentivize declaration and reduce the cost of undeclared bot traffic.

The policy model

  1. Kind of visitor. Undisclosed, person, or bot. The kind decides the surface: pages for a person, WebMCP functions for a bot.
  2. Access point. A tier of content for a person (free content, all content), or a function for a bot (search, get_latest, and so on).
  3. Requirement. Each access point has exactly one requirement: free, payment, or work. A function that accepts two kinds, for example pay or work, has two paths, one for each.

A requirement is a card that holds exactly one item, such as "Subscription, $1.00" or "Check 3 claims, size M". Two access points can point at the same card. For example, a person's "All content" and a bot's subscribe both point at the Subscription card. A card never lists more than one item, so each item stays something that can be configured, changed, and counted on its own.

People and bots are treated the same way: each gets access, and each access has one requirement.

Who pays

Payment is settled by the wallet, from the visitor's own funds or from a sponsored allowance. An institution can fund access for a student, or for a journalist, with a spending limit that can be capped for one agent and topped up for another. The site only sees a payment that settles. It doesn't need to know who funded it, so a student or a journalist needs no special case in the policy. The details of allowances belong to the mandate illustration (1c).

The wallet's part

The wallet supplies the proof of personhood, humanity, or liveness. The visitor doesn't need to do anything on the site. Because the wallet holds the proof, it can stay valid for a period the owner sets, so it isn't repeated site by site. The owner chooses which proof to require and how long it stays valid.

The card types

Card What it carries In the illustration
Disclosure What the visitor declares, and which proof is required Undisclosed, person, bot
Outcome A result with no further steps Rate-limited
Person Access points for a person, by tier Free content, All content
Bot Access points for a bot: its WebMCP functions and what each accepts search, get_latest, subscribe, spend_article_credit
Free One item: no exchange No exchange
Payment One item: an amount, settled by the wallet Subscription ($1.00), Per call ($0.02), Per article (1 credit)
Work One item: a sized task with a description Check 3 claims (M), Tag 5 photos (S)

What an owner would do

In a real builder, an owner would add a step, pick a proof and its validity period, and attach a requirement to each access point. Enforcing it at the edge means the gate in front of the site checks the visitor's proof before anything else is served. The illustration shows the result and not the editor: the edit icons and "Add step" rows on the cards are only styling.

What the illustration leaves out

Open questions

The gate

Who manages the gate: the site, a CDN (e.g. Cloudflare), or a third party (e.g. Mozilla)? A site running its own gate keeps control but has to build and maintain it. A CDN already sits in front of the traffic and can enforce at scale, but then the site depends on it for policy. A shared third party could make rules consistent across sites, at the cost of one more party that sees the traffic.

Proofs

Which proofs count (personhood, humanity, liveness), and can one be reused across sites without tracking the visitor? Reuse is what makes the wallet worth having, because the visitor proves it once. But a proof that is identical everywhere can also act as a tracking identifier. The design needs a way to show the proof is valid without handing every site the same token.

False declarations

How is one handled, such as a bot claiming to be a person? The incentive to declare only works if declaring falsely doesn't pay. Options include liveness checks that are hard for software, behavioral signals at the gate, and penalties that fall on the wallet that issued the false claim. Each trades some privacy or friction for accuracy.

Undisclosed

What exactly does an undisclosed visitor get: how limited, and is it the same for every site? A strict limit pushes visitors to declare, but it also catches legitimate visitors who can't or won't, such as people without a wallet. Each site will want its own balance, and a standard default would make behavior predictable for visitors.

Issuers

Who counts as a trusted issuer of humanity proofs and of sponsored allowances? A site has to decide which issuers it accepts, and that is a trust question that someone has to maintain, like a list of certificate authorities. A role for a neutral party such as Mozilla is one possibility. The alternative is each site keeping its own list, which is flexible but hard for visitors to predict.

Allowances

How is an allowance capped for one agent and topped up for another? A sponsor wants limits so one agent can't spend a whole budget, and wants to be able to move funds to where they're needed. That means the allowance needs a cap per agent and a way to re-fund it without issuing a new credential. The mandate illustration (1c) is where these budgets are shown.

A shared format

Should there be a standard format for these policies, so a bot can read them the same way on every site? A shared format would let a bot discover what a site accepts without custom code for each one, much as a standard certificate format does for HTTPS. The cost is agreeing on it, and a format that is too rigid may not fit the policies sites actually want.