NEWS · NOVEMBER 5, 2026 · AGENTIC AI

Cloudflare gives site owners per-agent visibility and rules for personal AI agents

On November 4, 2026 Cloudflare published new analytics that show personal AI agent visits as their own traffic category, plus access policies that can be set agent by agent. A second post the next day announced the Trusted Agent Delegation coalition, which Cloudflare says is meant to bring unresolved agent problems to the IETF.

01 · WHAT HAPPENED

Two Cloudflare posts in two days: tools first, a coalition next

On November 4, 2026 Cloudflare published "See the agents, set the terms: new visibility and control for personal agents", written by Marina Elmore and Nicole Justus. The post describes a personal agent as the go-between that turns a person's instructions into requests on websites. A customer might ask one to compare products, find support articles or complete a purchase, and to the site owner all three look like automated traffic. Cloudflare argues that without knowing which agents visit and what they do, it is hard to tell whether that traffic helps or harms, while blocking every agent can mean losing customers. The company says it is launching the new analytics and finer-grained access policies with that post.

The next day, November 5, Bryan Becker's post "Agents aren't humans (but they kinda are)" appeared. It is a long framing document arguing that a site can no longer identify, serve or hold to account the user standing behind an agent, and it announces the Trusted Agent Delegation (TAD) coalition. It also lists the previous day's tools among three steps Cloudflare customers can take as of that week, and describes them as already in production.

02 · DETAILS

Visibility, per-agent policy and account-level views

In Security Analytics, the Bot Analysis chart gains a Personal Assistants category beside Verified Bot, Likely Human, Likely Automated and Automated, so teams can isolate personal agent traffic over time. A new Personal Assistants widget under Top statistics lists each visiting agent and its request count. Filtering on one agent updates the other widgets to show which paths it asks for, which status codes come back and which networks it uses. Cloudflare says this reveals whether an agent is logging in, reading product pages, reaching checkout or running into errors. Policy controls sit in Security Settings under a section now named AI Crawler and Agent access policies, which expands the Agents setting Cloudflare introduced this summer. Owners set an overall Agents policy, then pick Allow (do not block), Block on pages with ads, or Block for each supported personal agent. Search and Training policies are unchanged. When a new zone is added, onboarding asks whether the site monetizes pages with ads and tailors the suggested policies to the answer.

Account Abuse Protection, which Cloudflare describes as currently in early access, adds agent-level views for sites that use it to guard their user accounts. A population overview shows which personal agents are active across users and how many accounts rely on them, every login and sign-up event shows whether an agent made it and which one, and fraud teams can narrow the account list to those with AI assistant activity. If an attacker appears to be using an agent on one account, an analyst can act on that one account through a WAF rule keyed to its Hashed User ID, leaving the agent available to everyone else. The second post adds that Business Insights, offered to Enterprise Bot Management customers, compares how much each operator crawls with the traffic it sends back.

03 · WHY IT MATTERS

Telling agents apart instead of blocking them all

Cloudflare's own summary of the benefit, in the November 5 post, is that "Unexplained automated traffic becomes traffic you can reason about." The larger argument in that post is that an agent is an intermediary: the site sees the agent rather than the user, and one agent can act for many users. The post lists friction this causes, among them account sharing, friendly fraud, resource exhaustion and broken per-user analytics, and says the site usually bears the cost. It also argues that agent traffic may overtake traditional human browsing in the near future, at which point simply blocking intermediaries stops working.

The TAD coalition is described as a place to bring representatives from across the ecosystem together, examine existing protocols and take well-defined problems to the IETF as needed. Cloudflare says it plans side meetings at the IETF and workshops later. The post names Web Bot Auth as its preferred method for agents to identify themselves and mentions MOLE and x402 as related protocols. It lists agent builders, platform operators and commercial or compliance teams as groups it wants involved, and points to trustedagentdelegation.org for details. In short, the analytics and policies are live, Account Abuse Protection is in early access, and the coalition is at the discussion stage. All of this comes from Cloudflare's own posts, so these are vendor claims.

04 · TÜRKİYE

Neither post mentions Türkiye, so this section is commentary

Neither Cloudflare post mentions Türkiye, Turkish sites or any local rule. Nothing in them says whether Turkish agent operators or Turkish sites are covered differently from anyone else. What follows is therefore not a fact from the sources but UNALSOFT commentary.

Our reading is that sites running ads or taking orders will increasingly need to decide per agent rather than all or nothing, and that seeing agent traffic is the first step. The unknowns are worth listing. The posts do not give plan tiers or prices for the new analytics and policies, apart from naming Enterprise Bot Management customers for Business Insights. They give no general availability date for Account Abuse Protection, and they do not list which personal agents count as supported. We wrote about Cloudflare's bot approach earlier, in an August 8 piece.

The UNALSOFT view

For a business on Cloudflare, the practical order is look first, then decide. Before choosing Allow or Block for any agent, check which agents actually reach your product pages, checkout and login, and what status codes they get. A site that earns from ads and a store that wants agent-assisted orders will reach different answers, and per-agent settings make that possible. Both sources are Cloudflare's own blog, so treat the capabilities as vendor-described until you see them in your own dashboard. Our agentic AI work starts from that same question: which agents act on your site, and on whose behalf.

Do you know which AI agents are visiting your site?

A short conversation is enough to map agent traffic on your site and decide where to allow or restrict it.

Message on WhatsApp