Already Have a WAF? Preparing for AI-Powered Automated Attacks

AI-powered attacks can hide in valid requests. Learn how bot management complements WAF protection while reducing disruption to legitimate users.
DAEN's avatar
Nov 05, 2026
Already Have a WAF? Preparing for AI-Powered Automated Attacks

Key takeaways

Valid login and API requests can still be part of an automated attack. Bot management complements WAF inspection by connecting identities, sessions, and behavior across requests. Effective protection also requires server-side authorization and policies that preserve legitimate user access.

Can stronger WAF rules stop attacks that abuse legitimate features?

Many businesses rely on a web application firewall (WAF) to protect their websites and applications. Detecting and blocking malicious requests, including SQL injection and cross-site scripting (XSS), remains a basic part of application security.

AI-powered automation creates another problem: login attempts and data queries can use valid request formats while serving an attacker's purpose. An attacker may try credentials across multiple accounts, collect information by changing query parameters, or rotate IP addresses while continuing the same activity.

These requests can look much like ordinary traffic. Attackers can also use AI to automate vulnerability discovery and exploitation, reducing the manual work needed to investigate each target.

Businesses therefore need to examine who is making requests, how those requests fit together, and how the activity develops over time, even when individual requests appear legitimate.

Updating WAF rules has long been a standard response to new threats. For AI-powered automated attacks, businesses also need bot management capabilities that can identify abuse across requests and sessions. Major global content delivery network (CDN) providers offer dedicated bot management features alongside their other security services, reflecting this broader need.

Recent incidents in Japan and South Korea show why these capabilities matter, and what businesses should assess in their own environments.

Related reading: WAF vs. Bot Management: Why Malicious Bots Get Through

1. AI-assisted vulnerability exploitation and API abuse in Japan and South Korea

Recent data breaches have put Japanese businesses on alert. In an October 2026 investigation, JPCERT/CC, Japan's computer security incident response coordination center, confirmed that attackers had used AI to scan for known vulnerabilities and conduct attacks through APIs. Investigators also found evidence that attackers had analyzed publicly available Android applications to identify API endpoints and understand how the applications communicated with their servers. Read JPCERT/CC's investigation.

Japan's National Institute of Information and Communications Technology (NICT) identified a related technique called an App-to-Backend (A2B) attack. Attackers analyze a publicly available application, then use what they learn to target its backend. The application itself does not need to contain a vulnerability: weaknesses in the server's authentication, authorization, or input validation can expose data. NICT also warned that AI could make application analysis and vulnerability discovery more efficient, allowing these attacks to spread. Read NICT's advisory.

South Korea faces similar risks. CrowdStrike reported a campaign against South Korean financial institutions involving ARTEX, an AI-assisted penetration testing tool, and large language models. Reported breaches included a loan inquiry service and an employee mobile work-support system. These incidents show why security reviews must cover internet-facing business systems and data lookup functions as well as customer websites. Read CrowdStrike's analysis.

ARTEX, Cairn, and CyberStrikeAI are among the AI-assisted penetration testing platforms that JPCERT/CC has identified as being misused in attack attempts. Tools intended for authorized security testing can be repurposed against customer websites, mobile app APIs, partner portals, and externally accessible employee systems. See JPCERT/CC's updated alert.

2. What makes AI-powered automated attacks different?

Treating automation as nothing more than the same request repeated at high speed leaves gaps in detection.

If access to one endpoint is denied, an attacker may try another API, change connection details, or adjust the interval between requests. The next request may look different from the last while still advancing the same objective.

Detection needs to account for that continuity. Check whether:

  • Related sessions and similar API call patterns persist after an IP address changes.

  • Data queries accumulate over long periods, even without a sudden traffic spike.

  • Requests reach protected APIs in a sequence that differs from the expected user journey.

  • Login activity reveals suspicious patterns across the number of accounts targeted and the success or failure of each attempt.

You do not have to prove that an attacker used AI before responding. Attack tools will continue to change, but the behavior needed to achieve an attacker's objective can leave observable traces. Base your response on automated behavior and evidence of abuse within your service.

3. Valid requests can still be part of an attack

A wolf disguised as a sheep walks past a WAF guard into a server room.
Automated attacks can resemble legitimate requests and pass individual request checks.

Login, order tracking, and reservation lookup are functions that customers need. Attackers can also use them to test stolen credentials, take over accounts, or collect information.

Service function

Behavior to investigate

Signals to examine

Account login

Repeatedly submitting credentials across multiple accounts

Number of accounts targeted, login failure rate, and relationships between sessions

Order and shipment lookup

Making consecutive requests while changing order or tracking identifiers

Changes in the records being requested, cumulative request volume, and access permissions

Product and reservation lookup

Persistently calling a particular API

Expected user journey, intervals between calls, and request volume by endpoint

Customer and business data lookup

Calling APIs in ways that are not available through the user interface

Requesting identity, API access path, and server-side authorization checks

These are examples to investigate, not automatic grounds for blocking. Legitimate applications and partner integrations may also make direct API calls or send large volumes of requests. Compare observed activity with the expected behavior of each service to distinguish malicious automation from authorized use.

A valid request format or successful login also does not establish that every subsequent action is legitimate. Continue evaluating data access and API activity after authentication.

4. Why bot management belongs alongside a WAF

A BotManager agent exposes and restrains an AI wolf disguised as a sheep beside a WAF guard.
BotManager connects identities, sessions, and behavioral signals to identify automated abuse.

A traditional WAF inspects requests for attack patterns and controls traffic according to configured access rules and request limits. Those capabilities remain useful when a request contains recognizable signs of an attack.

Defending against automated abuse also requires analysis of requesting identities, sessions, and behavior accumulated across multiple requests. Bot management adds behavioral and AI-based analysis to help identify activity that individual request checks may miss.

Security component

Primary responsibility

WAF

Inspect requests for attack patterns, such as SQL injection, and enforce configured access rules.

Bot management

Detect and control automated abuse, including credential attacks, distributed bot activity, repeated data collection, and abnormal API call patterns.

Servers and applications

Fix vulnerabilities and enforce authorization for each user and the data they request.

This comparison focuses on traditional WAF request inspection. Actual capabilities vary by vendor, product, and configuration; some WAF platforms also provide advanced bot management features.

5. Capabilities and support to look for in bot management

A bot management solution should complement WAF inspection by connecting activity across requests and assessing whether it constitutes automated abuse. The following capabilities and support areas use STCLab BotManager as a reference.

Correlate distributed requests

Evaluate multiple signals, including sessions, TLS fingerprints, and API call patterns, even when IP addresses or browser characteristics change. This helps reveal activity that request counts for a single IP address may miss. A matching fingerprint alone should not determine whether traffic is malicious.

Validate API call sequences

Examine the order in which endpoints are accessed and the time between steps. Where applicable, dynamic URLs can help control URL reuse that falls outside permitted conditions. Policies should reflect the legitimate access patterns of browsers, mobile applications, and partner APIs.

Detect repeated queries and data collection

Analyze repeated calls to specific URLs and request volumes over different time windows. Establish normal usage criteria for the data or function being protected, such as customer records, orders, or reservations, and respond to activity that falls outside those criteria.

Detect credential attacks and analyze login outcomes

Address account attacks such as credential stuffing and password spraying. Integrating server-side login results allows analysis to include successful and failed logins alongside attempt counts and the number of accounts targeted.

Cover the client, CDN, and server layers

AI-powered attacks can reach a service through more than one route. Detection and enforcement should cover activity at the end-user client, traffic passing through the CDN, and requests sent directly to the origin server.

Adapt policies to the service and industry

AI makes it easier to create tools tailored to a particular target. Standard policies need to be supported by ongoing analysis of the ways an individual service or industry could be attacked. Look for support that can develop and update custom policies as those threats change.

Protect legitimate users during policy rollout

Evaluate protection by both the abuse it stops and its effect on legitimate customers. False positives that prevent logins or purchases, or disrupt partner API integrations, create immediate operational problems.

Policies should be introduced gradually, after the service architecture and normal user journeys have been reviewed. During initial onboarding, customers should receive guidance based on traffic analysis and proof-of-concept (PoC) findings. A practical rollout can follow these steps:

  1. Define what to protect and what to allow. Identify protected endpoints and required integrations. Customer browsers, mobile apps, partner systems, and internal automation each need appropriate usage criteria. Discuss service-specific requirements and expected exceptions with the customer.

  2. Analyze live traffic in detection mode. Review which requests are flagged and whether legitimate high-volume queries are included. Use the findings to tune policies and thresholds.

  3. Apply enforcement to validated endpoints. Introduce blocking or additional verification where the policy has been tested. Monitor changes in repeated queries and credential attacks, along with legitimate login and transaction success rates and reports of false positives.

  4. Address customer-specific scenarios. Use detailed interviews and analysis to develop additional policies for unusual workflows and emerging attacks.

6. Assess your readiness for AI-powered attacks

If you already operate a WAF, use these questions to identify areas that need closer attention:

  • Can you detect repeated failed logins against the same account when the source IP address changes?

  • Can you spot requests concentrating on a particular lookup API even when total traffic remains normal?

  • Does your protection cover mobile app APIs, partner portals, and employee systems as well as the customer-facing website?

  • Can you compare flagged requests with the expected behavior of legitimate applications and partner integrations?

  • Can you measure how blocking policies affect legitimate logins and transactions?

  • Can you observe and detect attacks across the client, CDN, and server layers?

If any of these questions are difficult to answer, review your exposure and the coverage of your current controls.

STCLab offers a complimentary assessment and report on automated attacks that resemble legitimate requests. We review your service architecture and the information available for analysis, then use detection findings to identify endpoints and policies that may need additional protection.

To assess your defenses alongside your existing WAF, contact STCLab about a BotManager assessment. You can examine suspicious automated traffic reaching your service and define the protection your environment needs.

Explore BotManager and contact STCLab

Assess automated attacks reaching your service.

STCLab offers a complimentary assessment and report to help identify endpoints and policies that may need additional protection alongside your WAF.

Request a BotManager Assessment

Frequently asked questions (FAQ)

Why add bot management when a WAF is already in place?

A traditional WAF inspects requests for recognizable attack patterns and enforces access rules. Automated abuse can use valid request formats, so detection also needs to connect identities, sessions, and behavior across requests. Actual capabilities vary by WAF vendor, product, and configuration; some platforms already include advanced bot management features.

Do we need to confirm that an attacker used AI before responding?

No. Base your response on automated behavior and evidence of abuse within your service. Useful signals include activity continuing after IP changes, queries accumulating over time, unusual API call sequences, and login outcomes across multiple accounts.

What does a BotManager assessment involve?

STCLab offers a complimentary assessment and report on automated attacks that resemble legitimate requests. The team reviews your service architecture and available information, then uses detection findings to identify endpoints and policies that may need additional protection.

Share article