Tim Ayling | VP of Cyber Security Solutions EMEA
More About This Author >
Tim Ayling | VP of Cyber Security Solutions EMEA
More About This Author >
AI is transforming how applications are accessed, used and attacked. As autonomous agents become part of digital business, organisations must rethink how they protect their applications and APIs.
For decades, application security has been built around a relatively straightforward assumption: most interactions with websites and applications originate from humans.
That assumption no longer holds.
According to the 2026 Thales Bad Bot Report, automated traffic accounted for 53% of observed web traffic in 2025, with malicious bots responsible for 40% of the total. Meanwhile, AI-enabled bot attacks increased 12.5-fold year on year.
These figures highlight a fundamental shift in how the internet operates. Applications are increasingly accessed not just by people, but by intelligent software capable of browsing, making decisions, interacting with APIs and executing complex tasks.
For application security teams, this presents a challenge: the same AI capabilities transforming digital business are also changing the nature of cyberattacks.
Automation has been part of the cybercriminal toolkit for years. Credential stuffing, vulnerability scanning, web scraping and account takeover attacks are nothing new.
What AI changes is the potential speed, scale and adaptability of these activities.
Historically, automated attacks relied heavily on predefined scripts. AI can help attackers generate variations, analyse application responses and adapt techniques with less manual intervention.
Thales blocked 17.2 trillion bad bot requests in 2025, illustrating the enormous scale of malicious automation.
But the challenge is more than attack volume. Increasingly, attackers can exploit legitimate application functionality, making malicious activity harder to distinguish from genuine customer interactions.
This is particularly relevant as organisations introduce AI agents capable of independently navigating applications and completing business transactions.
APIs are the connective tissue of modern digital business, connecting applications, cloud services, partners and increasingly AI agents.
Unlike human users, AI agents don't necessarily need a traditional website interface. They can interact directly with APIs to retrieve information, execute transactions and complete workflows.
The 2026 Thales Bad Bot Report found that 27% of bot attacks targeted APIs.
Meanwhile, the Thales 2026 API Security Token & Authentication Risk Report, analysing more than 32,000 API endpoints, found that nearly 40% exhibited multiple authentication-related risks.
These findings highlight a growing challenge.
An API request can be properly authenticated and technically valid while still representing malicious activity. Attackers can exploit excessive permissions, manipulate parameters or abuse legitimate business processes.
Protecting APIs therefore requires more than authentication and traditional vulnerability detection. Organisations need visibility into their API estate and an understanding of how those APIs are being used.
Agentic AI is also challenging the traditional distinction between good and bad bots.
Imagine a consumer asking an AI assistant to find a product, compare prices, check availability and complete a purchase.
The agent might generate hundreds of requests and interact with multiple APIs without the customer ever visiting the retailer's website.
From a business perspective, this could be a valuable transaction. Yet some of that behaviour might resemble scraping or other automation that security systems have historically blocked.
Now consider a malicious actor using similar capabilities to manipulate inventory, exploit promotions or abuse application workflows.
The technical characteristics of legitimate and malicious automation may overlap.
This creates a new challenge for AppSec: identifying automated traffic is no longer enough. Organisations increasingly need to understand whether its behaviour is authorised and appropriate.
The 2026 Bad Bot Report found that 21% of bot attacks targeted business logic, demonstrating why protecting the integrity of application workflows is becoming increasingly important.
The goal must be to enable legitimate AI-driven interactions while preventing malicious automation.
AI isn't just changing how applications are attacked. It's changing the applications themselves.
Organisations are embedding large language models into customer service platforms, enterprise applications and automated business processes.
These systems introduce new risks, including prompt injection, insecure tool invocation and excessive permissions.
Consider an AI assistant that retrieves information and interacts with enterprise systems. If malicious content influences the agent's behaviour, it could attempt to invoke APIs or perform actions outside the user's original intent.
The attack may move across several components: untrusted content, the AI model, connected tools and underlying APIs.
The 2026 Thales Data Threat Report found that 61% of surveyed organisations reported their AI applications were being targeted.
For AppSec teams, this means protecting the model alone isn't sufficient. Security must extend across the entire AI-powered application workflow.
Traditional application security technologies remain essential, but increasingly they need to operate together.
Web Application Firewalls protect against exploitation attempts. API security provides visibility and protection against API abuse. Bot management addresses malicious automation, while DDoS protection maintains application availability.
AI application security adds another important capability, addressing threats such as prompt injection and unsafe agent interactions.
The challenge is that these technologies have often operated independently, creating gaps in visibility and protection.
An attack might begin with automated reconnaissance, progress to API abuse and ultimately exploit application business logic. Each individual security control may only see part of the activity.
The future of AppSec lies in using these capabilities together as one, sharing intelligence and applying coordinated protection across applications, APIs and automated interactions.
And because enterprise applications operate across cloud, on-premises and hybrid environments, that protection must be delivered consistently wherever applications run.
Security should adapt to the application, rather than requiring the application to adapt to security technology.
AI is changing the internet in two fundamental ways: increasing the capabilities available to attackers while enabling entirely new forms of legitimate automation.
For businesses, both developments matter.
Organisations must defend against increasingly sophisticated threats without preventing the AI-driven services and customer experiences that will support future growth.
That requires an integrated approach to application security, combining WAF, API security, bot management, DDoS protection and AI application security.
The objective isn't simply to block more traffic or deploy more security tools. It's to understand and protect the increasingly complex interactions taking place across digital businesses.
In the age of agentic AI, application security must evolve from protecting individual requests to protecting the integrity of entire digital interactions.
Because the future of Application Security is about understanding what connections to your applications are trying to do — and ensuring they can only do what they should.