If you run an online store, close to half of the non-human traffic hitting it today is an AI bot, not a search-engine crawler. That’s the headline number from Akamai’s latest State of the Internet security report, and the more important finding sits one layer beneath it: most merchants aren’t blocking that traffic, they’re just watching it go by.

The number

Akamai’s report, “Securing the Agentic Storefront: Attacks on Commerce” (published July 15, 2026), found that AI bots accounted for 47.9% of all commerce traffic across Akamai’s global network as of December 2025 — nearly half of every non-human request a store receives now comes from an AI system of some kind (Akamai newsroom, “Commerce Becomes the Epicenter for AI Bot Attacks and Agentic Fraud in 2026”).

Training crawlers make up the bulk of it: AI training bots drive more than 70% of AI bot triggers in commerce, and Akamai names OpenAI, ByteDance and Anthropic as the three most-observed operators. That’s a different traffic class from the shopping agents this site usually covers (ChatGPT, Claude, Perplexity acting on a shopper’s behalf) — it’s crawlers pulling product pages into training data, hitting stores at a volume most merchants have never had reason to measure.

The gap: monitored, not managed

The more consequential finding is how stores are handling all of that traffic. Akamai reports that commerce organizations put more than 90% of their AI bot activity into a “monitor only” category — logged, but not acted on — and of the remainder, roughly three-quarters is let through unrestricted. Combine the two and a large majority of AI bot traffic on a typical commerce site faces no access control at all, regardless of what that bot is actually doing once it’s in.

That gap is what’s opening the door to the fraud patterns Akamai flags alongside the traffic numbers:

  • Agent hijacking — attackers compromise a legitimate AI shopping assistant session to abuse stored payment credentials, rather than attacking a merchant’s site directly.
  • “Frankenstein” account fraud — attackers use LLMs to blend real stolen data (like a genuine Social Security number) with fabricated details to build synthetic identities cultivated to look like long-standing loyal customers, specifically to slip past static fraud rules.

Both rely on the same weakness: a bot-management setup built to log AI traffic, not to distinguish a legitimate crawler from an agent acting with malicious intent.

Why “block everything” isn’t the fix either

Akamai’s own recommendation isn’t a blanket lockdown — it’s replacing the binary allow/block model with risk-based governance that categorizes bots by intent and business value. That matters for agent-readiness specifically: a store that reflexively blocks all AI user agents also blocks the legitimate shopping agents (and training crawlers that drive future discoverability) it wants to let through. The fix isn’t less access, it’s knowing which bot is which before deciding.

That’s the same problem this site has covered from the identity side: telling a genuine agent apart from something spoofing its user-agent string is exactly what Web Bot Auth’s cryptographic signatures are built to solve, and it’s a prerequisite for the kind of terms-of-service and access policy work covered in Amazon v. Perplexity — neither works if a store can’t first tell which bots are calling in the first place.

What to check on your own store

  • Pull your own AI bot traffic share. If you’re on a CDN/WAF that categorizes bot traffic (Akamai, Cloudflare, or similar), check what fraction of your non-human hits are AI-labeled and what action policy is actually attached to each category — “monitor” is not the same as “reviewed.”
  • Separate crawlers from agents. A training crawler reading your product pages and a shopping agent trying to add an item to a cart on a shopper’s behalf need different rules; treating both as one undifferentiated “AI bot” bucket is how legitimate shopping agents end up rate-limited or blocked.
  • Audit account-security assumptions against LLM-assisted fraud, not just bot volume — Akamai’s “Frankenstein account” pattern is built specifically to pass identity checks tuned for older, cruder fraud signatures.

Bot volume and blocking policy are an access-layer question — a prerequisite, not a substitute, for the harder one AgentReady’s scan checks: once a legitimate agent gets through, can it actually find, understand and complete a purchase on your store.

Sources