Benjamin Fabre is the cofounder and CEO of DataDome.
For years, bot defense was simple. You stopped bad actors at the door and let real customers through. That worked when bots were mostly crawlers that scraped your homepage and moved on.
But that’s not the case anymore. According to my company’s “State of Bot & Agent Security Report, 2026,” “AI agents generated 605.6 million requests to high-risk endpoints” across our customer base in just the first half of 2026. We’re not talking about home pages or blog posts. AI agents are showing up on login pages, checkout and payment flows and account-creation forms, following the same paths as your customers.
From Crawlers To Customers
AI traffic used to be passive and operated much like a search engine crawler. Now AI agents complete the same multistep actions that used to require a person at a keyboard.
Monthly AI bot requests to login pages grew more than eight times in six months. Login pages made up 51.7% of all high-risk AI traffic in the first half of 2026, up from 23% a year earlier. That means your login page, not your homepage, is now the busiest AI entry point into your business.
The Authentication Problem
Login pages exist for a reason. They guard order history, saved payment methods and loyalty balances. A legitimate AI shopping assistant checking on a delivery has to log in somewhere, and that’s a valid use case.
But the same login page is just as attractive to a fraudster attempting credential stuffing or account takeover. The action looks identical, but the intent isn’t. That jump in login traffic could be legitimate agentic commerce, credential attacks or both at once. Block everything, and you shut out your own customers’ tools. Wave everything through, and you hand attackers an opening. Neither works.
Behavior, Not Identity
You can’t trust a bot based on what it claims to be anymore. According to my company’s report, “AI agent spoofing increased by 45% between February and July 2026.” Spoofing of user-prompted agents (the live assistants acting for real people in real time) quadrupled in that window. A bot claiming to be ChatGPT-User or PerplexityBot may very well be neither.
So the decision on whether to allow, throttle or block a visitor can’t rest on identity. It has to rest on behavior, what pages it touches, how fast it moves and whether the pattern matches a real agent or an attack.
The Cost Of Getting This Wrong In Both Directions
Tightening controls across the board may feel safe, but it isn’t free. If you block indiscriminately, you also block legitimate AI agents completing real purchases on behalf of paying customers. That’s lost conversions, right as agentic commerce is growing. Adobe found AI-driven traffic to U.S. retail sites grew 138% year over year in May 2026, converting 54% better than other sources.
The opposite mistake costs just as much. If you assume AI traffic is safe by default (or don’t have the visibility to check), you leave the door open to credential abuse, payment fraud and inventory manipulation. Endpoints gaining high-risk AI traffic like cart, payment and account creation connect straight to purchases and stored value, and getting it wrong shows up on the P&L fast.
What A Better Posture Looks Like
Start with visibility. Know what share of your login traffic is AI-driven. Without that, you’re making decisions blind on one of the biggest shifts in how your site gets used.
Then build behavioral context at the session level. What did this visitor touch before the login attempt? How quickly? Does the pattern match any known legitimate agent profile, or does it match a known attack signature? That’s what separates blocking the right traffic from blocking everything.
The perimeter has moved. It’s not the front door anymore; it’s every login, every form and every checkout on your site. Defenses built for doorstep bots aren’t built for 2026.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?