The wrong question is “Who are you?”
When platforms worry about bots, fraud or synthetic users, the obvious response is to collect more identity. Ask for a legal name, government document, face scan, phone number, address or payment instrument and bind the account to those records. That can be appropriate in regulated contexts, but it is a poor universal answer for the autonomous internet. Most online actions do not require a service to know a person's full civil identity. They require confidence about a much narrower fact.
A discussion forum may only need evidence that an account is backed by a human under an accepted uniqueness mechanism. A marketplace may need to know that a purchase agent is authorized by a legitimate account. A voting or community system may need one-person-one-action without learning who the person is. The design problem is therefore not simply identification. It is selective proof: proving the fact necessary for the decision while avoiding unnecessary disclosure.
Proof of human is a capability, not a universal passport
Permission Zero treats human proof as one possible root of authority, not as a demand that every person carry one global identity credential. Different environments can establish a human root differently. A consumer application might use a privacy-preserving uniqueness proof. An enterprise could rely on an internal identity provider. A financial application may combine regulated identity checks with a local account credential. What matters to the protocol is that a relying service can evaluate a proof and attach policy to it.
This avoids creating a single central honeypot of identity data. The trust layer can accept proof from multiple mechanisms while exposing only the attributes required for the current policy decision. The architecture becomes more resilient because a compromise or policy failure in one identity source does not automatically become a compromise of the entire network.
Minimum disclosure should be the default
The principle of minimum disclosure is straightforward: if a service needs to know that a user is over a threshold, do not disclose the exact value; if it needs evidence that a human satisfies a uniqueness policy, do not disclose a legal name; if it needs to know that an agent has valid authority, do not reveal unrelated permissions. Every additional attribute shared with a relying service expands privacy risk and creates more data that must be secured, retained, governed and eventually deleted.
A good authority system should therefore express proofs as narrow claims. The service receives the answer it needs and a way to verify that answer. The underlying evidence can remain with the appropriate issuer or proof system. This is particularly important in an AI-rich internet because agents may interact with hundreds of services on a person's behalf. Repeating full identity disclosure at every boundary would create a persistent map of a person's digital life.
Human proof and agent authorization are different events
A proof that a human exists should not automatically authorize an agent to act. These are separate steps. First, the system establishes a human root or another legitimate authority source. Second, that authority delegates a defined capability to a specific agent. Third, the relying service checks the delegation when the agent attempts an action. Keeping these events distinct limits damage if an agent credential is stolen or if a user later changes their mind.
This separation also makes revocation meaningful. A human proof may remain valid while one particular agent loses permission. A person can replace an agent, reduce its spending limit, change the approved counterparties or terminate the delegation without re-proving their entire identity. The authority relationship becomes dynamic instead of being fused permanently to a credential.
Privacy and Sybil resistance have to coexist
Systems that reward participation or enforce one-person-one-action face a difficult trade-off. If they collect too little information, one operator may create thousands of synthetic identities. If they collect too much, the system becomes invasive. The answer is not to pretend the trade-off disappears. The goal is to use cryptographic and behavioral techniques so the service can reject duplicate or obviously coordinated identities while learning as little as practical about the people involved.
Permission Zero's Human Layer is designed to sit at this boundary. It can combine proof-of-human inputs, local policy, rate controls and later reputation signals. No single technique should be treated as magical. High-value actions can require stronger proof than low-risk browsing. New accounts can begin with conservative limits. Suspiciously coordinated populations can be evaluated by the Swarm Defense layer even if each account separately presents a valid-looking credential.
Unlinkability matters
A privacy-preserving proof loses much of its value if every use produces the same globally recognizable identifier. That turns a pseudonymous credential into a tracking beacon. Where the application permits it, proofs should be scoped so one service cannot trivially correlate a person's activity across unrelated services. The service should learn what it needs for its own policy, not acquire a universal key for observing the person's entire digital history.
At the same time, local accountability can still exist. A relying service may need a stable pseudonymous handle within its own domain to enforce abuse limits, maintain reputation or investigate disputes. The architecture should support both requirements: local continuity where justified and cross-service unlinkability where broader correlation is unnecessary.
Recovery, revocation and lost credentials
Human-rooted authority also has to survive ordinary failure. People lose phones, change devices, rotate keys and make mistakes. A proof system that is private but impossible to recover from will not work at consumer scale. Recovery mechanisms need to be designed so they do not silently reintroduce a central administrator with unlimited power. Depending on the environment, recovery may involve multiple factors, trusted devices, organizational policy, social recovery or a fresh proof process.
Revocation should be equally explicit. If a credential or agent is compromised, relying services need a way to learn that the previous authority is no longer valid. The protocol should make expiration and revocation first-class fields rather than afterthoughts. Short-lived delegations reduce the amount of trust that survives unnoticed after a compromise.
The service should see a decision, not a dossier
The ideal relying-service experience is simple. The application sends a policy question: does this actor satisfy the conditions for this action? The trust layer evaluates the relevant human proof, agent identity, delegation and context, then returns a narrow decision and an auditable reason. The application does not need to receive a complete identity file in order to enforce that decision.
This model is especially important as AI agents interact programmatically. A service may process thousands of delegated actions without a human visiting a page. The privacy boundary therefore has to be enforced by protocol, not by a promise that a user will notice what data is being collected.
Prove enough, reveal less
The Human Layer is not an argument for anonymity at all costs, nor is it an argument against regulated identity where the law or risk requires it. It is an argument against making maximal identity exposure the default answer to every machine-era trust problem. Different actions deserve different levels of proof, and services should ask only for what they need.
Permission Zero's position is that a safe autonomous internet begins with human sovereignty over disclosure and delegation. A person should be able to establish a legitimate root of authority, authorize software to act, and prove that authority without automatically broadcasting a permanent identity trail. The strongest system is not the one that knows the most about everyone. It is the one that can make the right decision with the least unnecessary information.