OpenAI Operator, Deepseek, or even Alibaba Qwen, the rise of customizable AI agents (CUAs – Customizable User Agents) is disrupting the cybersecurity landscape. Very practical, they allow the automation of repetitive tasks on browsers or systems without human intervention, such as automatically filling out online forms, interacting with web interfaces, or retrieving and analyzing real-time data. Thanks to their flexibility, they are used in sectors such as customer support, data management, and software testing automation. However, this same flexibility makes them vulnerable to misuse by cybercriminals to carry out sophisticated identity-based attacks. This includes automated Card Testing attacks, which use stolen credit cards to test their validity before carrying out large-scale fraud, or Credential Stuffing attacks, which use lists of stolen credentials to gain access to online accounts.

When AI Becomes a Cybercriminal Weapon

These agents automate attacks, notably by exploiting APIs, browser automation tools, or advanced AI scripts. The automation of identity-based attacks follows five key steps:

1. Reconnaissance: Attackers use AI to scan SaaS services such as Microsoft 365 and Google Workspace, identifying active accounts and vulnerable entry points (OWASP A06:2021 – Exploitation of Vulnerable Credentials).
2. Initial Access: Thanks to AI, cybercriminals optimize credential stuffing and targeted phishing techniques. Credential stuffing involves using lists of stolen credentials (usernames and passwords) to attempt access to various online services. Since many users reuse their passwords, attackers automate login attempts to compromise accounts (MITRE ATT&CK – T1110.001). By analyzing victims’ communication patterns, they generate highly personalized phishing emails, bypassing traditional filters.
3. Persistence: Once an account is compromised, the attacker establishes persistent sessions or creates malicious OAuth applications (MITRE ATT&CK – T1556.003), ensuring stealthy and prolonged access.
4. Lateral Movement: AI allows attackers to map internal access points and extend the compromise to other strategic resources such as databases or cloud infrastructures (MITRE ATT&CK – T1021.002).
5. Exfiltration: Finally, AI agents facilitate the rapid and discreet extraction of sensitive data (MITRE ATT&CK – T1567.002), which can be sold on the dark web or used for ransomware attacks.

Credential Stuffing Attacks

In January 2024, a financial services company suffered a Credential Stuffing attack targeting its customer accounts, facilitated by an AI-powered bot. Over 200,000 login attempts were detected within 24 hours, resulting in 3,000 compromised accounts and an estimated $2.5 million financial loss (Cybersecurity Ventures Report 2024).

In March and April 2024, the streaming platform Roku experienced two separate Credential Stuffing attacks, leading to the compromise of 591,000 customer accounts in total.

These incidents highlight the critical need for robust security measures to defend against Credential Stuffing attacks.

How Does a Credential Stuffing Attack Work?

• Credential Collection:
  Attackers likely obtained stolen credentials from previous data breaches or dark web marketplaces. These credentials include username-password combinations commonly used across multiple online services.
• Automation Using an AI Bot:
  An AI bot was deployed to automate large-scale login attempts. It systematically tested stolen credentials across the financial services company’s systems, exploiting the fact that many users reuse passwords across multiple platforms.
• Detection & Account Compromise:
  By leveraging automation, the bot executed a high volume of login attempts in a short time, maximizing the probability of success. Once accounts were compromised, attackers gained access to sensitive information and could initiate fraudulent transactions.

As AI-driven attacks continue to evolve, organizations must implement advanced security controls to mitigate Credential Stuffing threats effectively.

How to Detect a Credential Stuffing Attempt?

Early detection relies on advanced tools and continuous monitoring:

• Increase in failed login attempts: A sudden spike in unsuccessful login attempts indicates an automated attack.
• Access from multiple countries in a short time: Attackers use bots from various locations to mask their activity.
• Increased bot activity: Detection of extremely fast login attempts or CAPTCHA bypass through network traffic analysis.

How to Respond to a Credential Stuffing Attack?

A rapid and effective response is essential to limit damage and protect user accounts:

• Lock compromised accounts: Identify and immediately suspend affected accounts. Require a password reset before restoring access.
• Monitor suspicious activities: Analyze unusual behaviors such as login attempt spikes, access from unknown IPs, or abnormal transactions.
• Notify affected users: Alert victims in real-time and recommend enabling multi-factor authentication (MFA).
• Strengthen security measures: Implement IP blacklists, CAPTCHA, and rate limiting to restrict login attempts.

Best Practices to Prevent Credential Stuffing Attacks

• Multi-Factor Authentication (MFA): Add an extra validation step (OTP, biometrics) to block attackers even if they have valid credentials.
• Rate Limiting: Restrict the number of login attempts per IP address to slow down attacks.
• CAPTCHA: Deploy anti-bot tests to prevent large-scale automated login attempts.
• Strong Password Policies: Require complex passwords and encourage regular password updates.
• Anomaly Detection & Monitoring: Continuously analyze login patterns and suspicious behavior to detect attacks in real-time.
• Advanced approaches to strengthen defense: Attack detection can be enhanced through AI-based Threat Hunting solutions, such as User and Entity Behavior Analytics (UEBA), which help identify abnormal behaviors and suspicious logins. The adoption of passwordless authentication with FIDO2/WebAuthn completely eliminates the risk associated with stolen credentials. Finally, real-time dark web monitoring, using platforms like Cybersixgill or Recorded Future, allows organizations to anticipate threats by quickly identifying credential leaks before they are exploited.

By combining these strategies, businesses can reinforce account security and mitigate the impact of automated attacks.