Automated traffic now outpaces human traffic by up to eight times, and AI agents have moved away from shopping — raising a security question many organizations are ill-equipped to answer.
getty
Internet traffic is set to grow eight times faster than human traffic by 2025. The most important change isn’t volume – it’s what automation is doing now.
For years, the bot problem was a distraction. Scrapers have taken price points. The crawlers picked up the content. Screenshots of information printed on login pages. Those are still real problems. But the nature of traffic has changed, and the security considerations of many organizations have not yet caught up.
AI agents are no longer just browsing the web. They are still working on it.
A Different Type of Automation
A new benchmark report from Human Security, which analyzed more than a quadrillion transactions for its customers by 2025, puts the numbers in flux. Monthly AI-driven traffic increased by 187% from January to December. Agent AI traffic—systems that search, fill out forms, manage accounts and make purchases entirely on behalf of users—increased 7,851% year-over-year.
The AI agent that fills out the payment is not just reading. Makes financial decisions on behalf of the individual user, interacts with payment systems and account tools. The security implications are very different from reading your product pages.
I had the opportunity to chat with Todd Thiemann, a cybersecurity industry analyst with Omdia, about what the change means for security teams. His plan was specific: “AI agents have the promise of improving performance and productivity, but that new information needs to be managed and protected for compliance reasons, for cybersecurity reasons and to help business growth.”
AI agents are not the only traffic generator to choose from. It’s a new group that can act, decide and commit—and most corporate identity settings aren’t designed with them in mind.
Wrong Question
Security teams have spent years asking the same question: is this traffic from a bot or a human? Creating that plan made sense when bots were often the antagonists and humans were legitimate. It doesn’t hold anymore.
An AI agent browsing product pages, logging into an account and completing a purchase does exactly what a sophisticated bot attack looks like. Behavior is the same in practice. The difference is the intent—and the intent does not appear in the user agent string.
Of all the interactions analyzed, only half of one percent separates harmful automation from harmful automation. Organizations that block all automation will turn away from legitimate business. Those who allow it to go unchecked are accepting fraud. The real question is not whether the traffic is automated – it’s whether the communication provided is credible.
Threat Actors Follow the Same Playbook
Threat actors are targeting the same areas where agency AI is active: product pages, account management and checkout. That meeting did not happen by chance.
Attempts to compromise after logging into an account will more than quadruple by 2025, totaling 402,000 per organization. Login defenses have improved so much that attackers now wait until after authentication, abuse time tokens and use limited-step systems instead of forcing their way through the front door.
Scraping attacks now make up almost 20% of internet traffic worldwide – almost double the figure by 2022. For highly targeted organizations, it’s more than 60%. The volume of the card has increased by 250% at the same time.
Researchers have already written AI agents that attack card cards – cycling through card additions and payment attempts with old browsers, showing the theft process established without manual effort. The same tools designed to help consumers shop appear to be equally useful for theft.
Published Identity is No Longer Trusted
The spoofing problem compounds this. Attackers impersonate well-known AI attackers — claiming to be ChatGPT, Mistral, or Perplexity bots — to exploit trust organizations that go by those names. Whitelisting based on user threads provides access to actors who are not who they claim to be. And a single company can use crawlers, scrapers and agency systems at the same time, so operator-level access decisions don’t map cleanly to behavior. Information dissemination is the starting point, not the answer.
The Building Gap
Tools built for the human-facing Web are not designed for this. Bot detection assumes that the traffic is legitimate by humans. CAPTCHAs and rate limits assume that people have a natural ceiling on request volume. None of those assumptions hold when a legitimate seller can read 200 pages of a product in a minute before making a purchase.
What’s needed is the ability to understand the purpose of each interaction and leverage trust throughout the entire life cycle – not just at the point of entry. That means knowing which agents are running, what they are allowed to do and which downstream actions have the right permissions.
Thiemann put the defender’s challenge clearly: “From a defender’s point of view, you need to think about human identity, non-human identity and then AI agents who can make decisions and take action. Organizations need to manage and secure AI agents to avoid data breaches, fraud and other evils, and they need to do it right to accelerate their businesses.”
Many organizations have not had that innovation. They are running the 2018 defense against the 2026 traffic.
Where This Goes Next
More than 95% of AI-driven traffic flows through shopping and e-commerce, streaming and media and travel and hospitality. OpenAI alone accounts for 69% of AI bot traffic. Organizations of those sectors already live in this environment. Early data from 2026 suggests that the pace has not slowed, and the strategic decisions they make now – who gets access, under what conditions, with what authentication – will affect risk exposure and revenue for years.
The internet has crossed the line. Most of the traffic is automated. AI agents buy things. Fraud follows the same conditions as legal automation. The security question has changed.
The tools that many organizations use do not have them.
#Internet #Built #People #Anymore