Half the Internet is no longer human. Does your fraud stack know which half it is talking

For most of the past decade, the playbook for automated traffic was simple: detect the bot, block the bot. That logic is about to break. According to figures cited by Socure, daily requests from AI agents grew by more than 1,700% over the year to September 2026, and for the first time more than half of internet traffic was not human. At the same time, Accenture’s 2026 Consumer Pulse report found that 85% of consumers are open to collaborating with an AI agent, and 74% are open to letting one execute specific tasks on their behalf.

Taken together, those numbers describe a quiet but profound change in who, or what, is knocking on the front door. The agent that tries to book a flight, open an account or dispute a charge may well be acting for a legitimate, valuable customer. It may also be acting for a fraudster. Most organisations today cannot tell the difference, and many cannot even see how many agents are interacting with them or what those agents are doing.

The accountability layer is missing
Socure, an identity and risk intelligence provider, launched Agent Intelligence this week to address precisely that gap. The core problem, as the company frames it, is that an agent inserts another layer between an action and the person accountable for it, while every verification method running at scale in production was designed to evaluate a human being.

Existing tools can flag automation, but they were never built to answer the questions a risk team now needs answered: is this an agent, which agent is it, and which verified human or business stands behind it?

Socure’s CEO Johnny Ayers argues that blunt tactics fail in both directions. Block every agent, and you turn away your next generation of customers. Let them all through, and you hand fraudsters an unguarded entrance. In his view, the way out is not better bot detection but a clear determination of agent, identity and principal.

Three layers, three timelines
The suite is built in three layers that arrive on different schedules, which matters for anyone planning around it. Agent Detection is generally available now, delivered inside Socure’s existing Digital Intelligence offering. It identifies and classifies AI agents driving browser sessions, whether or not they announce themselves, and names the agent when it can. Rather than a yes-or-no verdict, it returns a confidence score with reason codes, a sensible design choice given that control of a session can pass between a human and an agent partway through. Socure reports that named consumer agents, including Muse and Instinct, were identified with 99.99% accuracy across 700 behavioral variables, though that figure comes from controlled lab testing rather than production traffic.

Agent TrustScore, currently open to beta customers, layers judgment on top of detection. It combines three categories of evidence: identity, meaning who the agent is and who stands behind it; authorization, meaning what it is permitted to do and whether its behavior stays within those limits; and risk, meaning what the histories of the principal and operator suggest, including signs of replay or impersonation. Each is returned separately, so a bank and a retailer can weigh them differently according to their own appetite for risk.

Agent Credential, expected to reach general availability in the first quarter of 2027, is the most structurally interesting piece. It is a W3C Verifiable Credential, issued through SocurePass by an operator to its agent, which binds that agent to a verified principal in Socure’s identity graph. Any site can verify it without calling Socure, and because the format supports cryptographic chaining and proof of possession, integrity can follow a delegation chain when one agent hands work to another.

Why the graph is the pitch
Socure’s central argument is that agents should be treated as entities in an identity graph rather than anonymous sessions. The company says it resolves agents as nodes in the same global graph that already carries its consumer identifiers, drawing on more than 11 billion identity decisions annually. An agent then becomes a known entity attached to a known person, with history, and organizations can allow, step up or block according to their own thresholds. Platforms that deploy agents can, in turn, issue credentials that make their agents easier to identify and less likely to be blocked.

The company is also positioning itself against a fragmented Know Your Agent market, where some vendors classify automated traffic while others only establish an agent’s identity or verify delegated authority. Socure calls its offering the only one that detects, scores and credentials agents on a single identity graph, a claim CIOs should treat as marketing until they see it tested against their own traffic.

One design decision deserves credit. Agent Intelligence is deliberately agnostic to agent type and credential issuer, and Agent TrustScore evaluates credentials from any issuer as well as agents carrying none.