Bots are now most of the web: reading the 2026 Bad Bot Report for financial services

Every year, Thales (which acquired Imperva) publishes a report on automated traffic across the sites it protects. The 2026 edition, covering full-year 2025, crossed a line the industry had been approaching for years: more than half of the traffic in its dataset was not human.

This post pulls out the figures that matter to banks, fintechs and anyone running a site where money moves, and adds the caveats that should travel with them.

The headline figures

All six figures come from the report's summary page. Cloudflare, looking at its own network, reached a similar conclusion independently: in September 2026 it wrote that “for the first time, more than half of Internet traffic wasn't human” this year.

Three caveats before quoting them

1. It is a dataset, not a census. The figures describe traffic on sites Thales protects. Your own mix may be very different. “53% of the internet” does not mean 53% of your sign-ins.
2. Automated is not the same as harmful. The gap between 53% bots and 40% bad bots is search crawlers, monitoring tools and legitimate automation. Increasingly, it also includes AI agents doing real work for real people.
3. Shares are not losses. 46% of account-takeover attacks targeting financial services says where attackers aim, not how often they succeed.

A third category: AI agents

The most important change in this year's report is a definition. In a post accompanying the release, “Bad Bots in the Agentic Age”, Thales's Tim Chang wrote that the company has “had to introduce a third category of automated traffic: AI agents.” Agents blend into legitimate use, which makes them harder to classify, and he warns that the AI traffic that identifies itself is only a fraction of the total, because attackers can run models that don't.

The old model had two boxes: good bots you let in, bad bots you keep out. Agents break it. The same agent, on the same site, can be doing exactly what its user wants at 10:00 and something its user never intended at 10:05. Cloudflare made the same argument from the other side: “AI bot” is no longer a useful label, because “what matters is what a bot does.”

Why financial services takes so much of it

Thales attributes financial services' 24% share to the APIs behind online banking. In the blog's words, “bots bypass user interfaces (UI) and interact directly with backend services”, using authenticated requests that many web-layer controls miss. Money is the obvious motive. The FBI's 2025 Internet Crime Report recorded about $17.6 billion in cyber-enabled fraud losses, roughly 85% of all reported losses, The Record reported.

Account takeover is the specific threat the 46% figure points to, and it is the one where agents change the picture most. A bot that steals a password and an agent that was given one end up in the same place: inside a valid session, with valid credentials. Every login check passed in both cases. That is the scenario US banks described in October 2026, when Axios reported third-party agents operating inside customer accounts.

Agents at work, not just attackers

The rise in automation is not only criminal. Cloudflare reported that daily requests from AI agents on its network grew by more than 1,700% in a year. Inside companies, agents are already widespread: 86.4% of organisations in Salt Security's 2H 2026 survey operated at least one, and 49.8% confirmed or suspected one had taken an unexpected or unauthorised action, Campus Technology reported.

Put those together and the question a bank faces is no longer only “is this a bot?”. Many sessions will involve automation of some kind. The question that matters is narrower and more useful: at the moment of this action, is a person in control?

Implications

• Edge filtering still matters, and the report is a good argument for it. But a growing share of risky automation arrives with valid credentials, inside sessions that edge controls have already let through.
• The risk concentrates at actions. Withdrawals, payee changes, credential resets and transfers are where automation turns into loss. They deserve their own check, at the moment they happen.
• “Block all bots” is no longer the goal. Customers will use agents. The goal is to let agents help while making sure the steps that move money or change control of an account are taken by a person.

That last point is the one HEIP is designed around. It works alongside bot management and fraud tools, and answers the question they don't ask: at the moment of this action, is a person at the controls?