The report
Forter's updated October 8 report says AI agents were placing more orders on its merchant network, but the figures do not establish how many people worldwide shopped through agents. The vendor's observation window is August 1 to October 7, 2026. Its own classification separates commerce-protocol integrations from agents operating ordinary retail websites through screen scraping.
Read the denominators before the growth rate
| Forter-reported metric | What it measures |
|---|---|
| 24% August-to-September growth, or 48% including Instinct and Grok Bot | Two different cohorts of agent orders detected on Forter's network. |
| 5.3x early-August agentic order level | Relative detected network volume after Muse appeared, not fivefold growth in all online retail. |
| 229% month-over-month comparison in the week Muse was first detected; 46% of agentic orders in the final September week attributed to Muse | Change and share within Forter's observed traffic, not proof of a global launch date or market share. |
| Screen-scraping share approximately 1% to 65% | Changing mix within detected agent orders, not the share of all purchases on the internet. |
| $223 average agentic order; 27% of agentic orders $50 or less | Order-value statistics from the classified merchant-network sample. |
Forter says it first reliably identified Muse on September 21, and warns that detection can follow actual use. It also reports Muse activity at 74% of stores that received any observed agentic order, compared with 45% for ChatGPT Agent. These are coverage fractions of its measured stores, not a worldwide merchant census. The absolute global denominator, detector error rates and independently validated attribution criteria are not supplied.
Why the checkout mechanism matters for security
Forter groups OpenAI and Google protocol-based checkouts separately from browsing agents operating a merchant's existing checkout pages. A protocol may offer a direct place to represent a delegated mandate and identify the requester. A browser agent can use an already signed-in session, which creates ambiguity about whether the human authorized the specific merchant, item, amount and submission. Neither integration type proves valid consent by itself.
The Authority-Inference Separation test explores independent approval of exact transaction effects. The OpenAI Sepolia incident concerns use of funds without an assigned mandate. These records explain relevant trust boundaries, but Forter's telemetry does not link either mechanism to its observed orders or losses. Its primary report documents commerce activity, not a measured fraud or agent-compromise rate.
The defensible conclusion is narrower than the claim circulating in social media: agent-operated purchases are observable in one merchant network, with a rapidly changing mix of interfaces. Millions of distinct human buyers, successful delegated authorization and completed post-purchase obligations remain separate, unmeasured questions.