Research

Arkose Labs: State of Agentic AI Trust, Q3 2026

A padlock fastened on a gate latch
Photo: Squire Old English Padlock and Latch by Partonez, CC BY-SA 4.0, via Wikimedia Commons

Executive summary

Only 2.3% of legitimate AI agent sessions carried a verified Web Bot Auth signature; ChatGPT agent was "the only agent that signed at scale". 77.4% of AI agent sessions were classed as legitimate, versus 7.9% of known bot sessions.

Key findings

  • 97.7% Most legitimate agents are unverified
    Share of legitimate AI agent sessions with no verified Web Bot Auth signature.
  • 78.4% ChatGPT agent signs
    Share of its sessions that were signed.
  • 65.2% Agents are widespread on protected flows
    Share of customers who saw AI agents on protected flows on September 30.
  • 70.5% Registration sessions are mostly fake
    Share of registration sessions that were fake-account attempts.

Methodology

Arkose's own production traffic from July 1 to September 30, 2026, with test and internal accounts excluded, limited to five protected actions. Muse's pre-September 22 figures are reconstructed from stored data, and the Muse and Instinct profiles rest on eight hours of data.

Arkose Labs published the report on October 20, 2026, covering July 1 to September 30. PPC Land analysed it.

How to read it

Arkose sells fraud-prevention products that compete with others in this area, and the data is its own production traffic. Raw session counts, customer numbers and sample sizes are not published, and some reach figures compare a four-day average with a single day.

Sources

  1. PPC Land: 97.7% of legitimate AI agent sessions carry no verified ID, Arkose finds (opens in a new tab) News report ·

Get MarTech Roundup in your inbox

What changed in marketing technology, why it matters, and what to do next.

We will email you to confirm. You can unsubscribe at any time.